FE SCHOOL
j CALIFORNIA 03345

NAVAL POSTGRADUATE SCHOOL

Monterey, California

THESIS

AN

ASSESSMENT OF THE POTENTIAL

ROLE OF

MULTI SPECTRAL IMAGERY IN BATHYMETRIC

CHARTING

by

Richard Thomas Joy

September 1984

Thesis

Advisor: James

L.

Mueller

Approved for public release; distribution unlimited

T222195

SECURITY CLASSIFICATION OF THIS PAGE (Mrnan D.;» Entered)

REPORT DOCUMENTATION PAGE

READ INSTRUCTIONS
BEFORE COMPLETING FORM

1. REPORT NUMBER

2. GOVT ACCESSION NO.

3. RECIPIENT'S CATALOG NUMBER

4. TITLE (and Subtitle)

An Assessment of the Potential Role of Satellite
Multispectral Imagery in Bathymetric Mapping

5. TYPE OF REPORT A PERIOD COVERED

Master's Thesis
September 1984

6. PERFORMING ORG. REPORT NUMBER

7. AUTHORfsj

Richard Thomas Joy

8. CONTRACT OR GRANT NUMBERf*.)

9. PERFORMING ORGANIZATION NAME AND ADDRESS

Naval Postgraduate School
Monterey, California 93943

10. PROGRAM ELEMENT, PROJECT, TASK
AREA 4 WORK UNIT NUMBERS

II. CONTROLLING OFFICE NAME ANO ADDRESS

Naval Postgraduate School
Monterey, California 93943

12. REPORT DATE

September 1984

13. NUMBER OF PAGES

97

14. MONITORING AGENCY NAME ft ADDRESSf// different from Controlling Of/Ice)

IS. SECURITY CLASS, (of this report)

UNCLASSIFIED

15a. DECLASSIFICATION/ DOWNGRADING
SCHEDULE

16. DISTRIBUTION STATEMENT (of this Report)

Approved for public release , distribution unlimited

17. DISTRIBUTION STATEMENT (of the abstract entered In Block 20, It different from Report)

18. SUPPLEMENTARY NOTES

19. KEY WORDS (Continue on reverse aide II necessary and Identify by block number)

Satellite bathymetry, multispectral imagery, Landsat, Multispectral
Scanner (MSS) , remote sensing, bathymetric chart revision, Great Bahama
Bank, West Florida Shelf.

20. ABSTRACT (Continue on reverse side If necessary and Identify by block number)

Previous research has demonstrated the feasibility of deriving water depth
information from Landsat Multispectral Scanner (MSS) digital data. However,
previously published results, analysed together with two new case studies,
show that the magnitude of errors (approximately 1-2 meters) in MSS single-
band depth extimates is too large for direct production of bathymetric
charts. Better accuracy is possible, though, if MSS data are used to
interpolate conventional soundings between survey tracklines, especially
if the survey vessels obtain concurrent optical ground truth data.

DD

FORM
1 JAN 73

1473 EDITION OF 1 NOV 65 IS OBSOLETE

S, N 0102- LF- 014- 6601

SECURITY CLASSIFICATION OF THIS PAGE (When Data gntarad)

SECURITY CLASSIFICATION OP THIS PAGE t *h*n Omm Bntmr*d)

If depth accuracy standards can be met, the MSS interpolation approach will
be extremely cost effective. In addition, MSS imagery is shown to be a
useful tool for planning and managing conventional surveys. A recommended
set of procedures is outlined for incorporating MSS image data into an
operational bathymetric mapping program. A comprehensive program of
development and operational demonstration surveys is recommended to
convincingly establish the utility and cost effectiveness of these
procedures.

S/N 0102- LF- 014-6601

SECURITY CLASSIFICATION OF THIS PAGEfWtfn Dmtm Bnfr*d)

Approved for public release, distribution unlimited

An Assessment of the Potential Role of
Multispectral Imagery in Bathymetric Charting

by

Richard T. Joy

Defense Mapping Agency Hydrographic/Topographic Center

B.S., University of Maryland, 1976

Submitted in partial fulfillment of the
requirements for the degree of

MASTER OF SCIENCE IN HYDROGRAPHIC SCIENCES

from the

NAVAL POSTGRADUATE SCHOOL
September 1984

NAVAL _

MONTEREY, C .

(A t a
ABSTRACT -r-u, -»

Previous research has demonstrated the feasibility of
deriving water depth information from Landsat Mul t ispectral
Scanner (MSS) digital data. However, previously published
results, analysed together with two new case studies, show
that the magnitude of errors (approximately 1-2 meters) in
MSS single band depth estimates is too large for direct
production of bathymetric charts. Better accuracy is
possible, though, if MSS data are used to interpolate
conventional soundings between survey tracklines,
especially if the survey vessels obtain concurrent optical
ground truth data. If depth accuracy standards can be met,
the MSS interpolation approach will be extremely cost
effective. In addition, MSS imagery is shown to be a
useful tool for planning and managing conventional surveys.
A recommended set of procedures is outlined for
incorporating MSS image data into an operational
bathymetric mapping program. A comprehensive program of
development and operational demonstration surveys is
recommended to convincingly establish the utility and cost
effectiveness of these procedures.

A-

TABLE OF CONTENTS "^

03

I. INTRODUCTION 11

A. BACKGROUND 11

B. DEFENSE MAPPING AGENCY HYDROGRAPHIC/
TOPOGRAPHIC CENTER CHARTING PROGRAM 12

C. SCOPE AND FORMAT OF THESIS 16

II. WATER DEPTH ESTIMATION FROM MULTISPECTRAL
SATELLITE IMAGERY 18

A. THE LANDSAT MULTISPECTRAL SCANNER (MSS) 19

B. ALGORITHM FOR CONVERTING MSS RADIANCES TO

WATER DEPTHS...- 21

C. PREVIOUS RESULTS -. . . 30

D. SYNOPSIS 37

III. SELECTED CASE STUDIES 39

A. INTRODUCTION 3 9

B. SITE CHARACTERISTICS 39

C. DATA DESCRIPTION 44

1. Landsat Images 44

2. Water Depth Data 45

D. METHODS OF ANALYSIS 46

E. RESULTS • 49

IV. DISCUSSION 56

A. ANALYSIS OF RESULTS 56

B. ERROR IMPACT ANALYSIS 58

1. Introduction 58

2. Sources of Error in Depth Estimation 59

C. IMPACT OF ERRORS ON POTENTIAL USES 60

1. Chart Evaluation Using MSS Derived
Bathymetry 60

2. Use of MSS Derived Depths as a Pre-Survey
Planning Tool 61

3. Use of MSS Imagery as a Direct Source of
Bathymetric Data 63

4. Use of MSS-Derived Depths as an
Interpolation Tool 64

D. COST BENEFIT ANALYSIS 66

E. RELATED TECHNOLOGICAL DEVELOPMENTS 68

V. CONCLUSIONS AND RECOMMENDATIONS 71

A. RECOMMENDED CHART REVISION AND SURVEY PLANNING
PROCEDURES 72

B. BATHYMETRIC SURVEY ENHANCEMENT USING MSS DEPTH
INTERPOLATION 74

C. RECOMMENDATIONS 78

APPENDIX A - LANDSAT SATELLITE SERIES AND THE

MULTISPECTRAL SCANNER 81

1. LANDSAT SERIES - CHARACTERISTICS 81

2. LANDSAT MULT I SPECTRAL SCANNER 82

3. SIGNAL PROCESSING TO PRODUCE COMPUTER
COMPATIBLE TAPES 83

APPENDIX B - NAVIGATION OF LANDSAT SCENES 86

LIST OF REFERENCES 92

INITIAL DISTRIBUTION LIST 95

LIST OF TABLES

1. ERIM Study - Results Using Single Bottom
Reflectance Value 34

2. ERIM Study - Results Using Individual Bottom
Reflectance Values 35

3. Identification of Landsat-1 MSS Imagery and
Key Radiation Parameters for Case Study Areas

1 and 2 45

4. Sources Used for Control Depths: West Florida
Shelf Test Area 45

5. Results of Analysis for Test Areas 1 and 2 54

LIST OF FIGURES

1. Inadequacies in World-Wide Coastal Hydrographic
Data 14

2. Case Study 1 - Site 1 41

3. Case Study 1 - Site 2 42

4. Case Study 2 - Sites 3 and 4 43

5. Plot of Log Reflectance Anomaly (X) vs. D for

Site 1 Test Sample 50

6. Plot of Log Reflectance Anomaly (X) vs. D for

Site 1 and Site 2 Combined Test Sample 51

7. Plot of Log Reflectance Anomaly (X) vs. D for

Site 3 Test Sample 52

8. Plot of Log Reflectance Anomaly (X) vs. D for

Site 4 Test Sample 53

9. Use of Mul t ispectral Imagery in Chart Revision

and Long-Term Planning 73

10. Integration of Multispectral Imagery into
Hydrographic Survey Program 75

ACKNOWLEDGEMENTS

I would like to thank my advisor, Dr. James L. Mueller, for
his dedicated guidance throughout this thesis project. I
especially appreciate his being available for lengthy
"long-distance" consultations. I would also like to
express my gratitude for the assistance I received from
DMAHTC (in the form of data, references, and graphic arts
support). Finally I would like to thank M. Marlow and S.
Apolinario for dedicating so much of their time typing this
thesis.

10

I. INTRODUCTION

A. BACKGROUND

The compilation of accurate and reliable nautical
charts, in support of military operations and commercial
shipping interests, is dependent upon the acquisition of
complete and up-to-date hydrographic information.
Presently, however, less than 20 percent of the world's
oceans have been surveyed adequately enough to accurately
portray bottom topography (Hammack, 1977). The result is
that many potentially useful shipping lanes may be labelled
unnavigable simply because increasingly precious ship
survey time is unavailable for demarcation or confirmation
of safe passage routes. More importantly, the potential
for shipwrecks or groundings increases when certain
shipping lanes are not routinely re-surveyed. In addition,
there is increasing demand for highly accurate surveys over
potential new routes.

Existing hydrographic resources are unable to respond to
these requirements in a timely fashion. This situation has
prompted research into alternative techniques for acquiring
hydrographic information. One promising approach now being
investigated is water depth estimation from mul tispectral
satellite images measured in the visible wavelength part of
the electromagnetic spectrum.

11

The feasibility of deriving water depths from satellite
mul t ispectral data has been demonstrated for selected
areas. To obtain maximum benefits from this technique,
however, further research will need to be directed towards
the specific uses to which this kind of data can best be
put. To explore this topic, a careful examination of an
actual hydrographic charting operation was required. The
Defense Mapping Agency Hydrographic/Topographic Center
(DMAHTC) program was chosen as the subject of this study
because of its extensive role in the global charting effort
and because of its history of supporting satellite
multispectral bathymetric research.

B. DEFENSE MAPPING AGENCY HYDROGRAPHIC/TOPOGRAPHIC CENTER

CHARTING PROGRAM

DMAHTC is tasked with providing hydrographic and
navigational information to the Armed forces of the United
States, other Department of Defense components, the
Merchant Marine, and mariners in general (Defense Mapping
Agency Hydrographic/Topographic Center, 1980). To
accomplish this mission, DMAHTC tasks the Naval
Oceanographic Office (NAVOCEANO) with the ongoing
acquisition of survey data. In addition, DMAHTC receives
data from other countries as part of an international
exchange program. Concerned mariners also supply a

12

significant amount of information in the form of sounding
data and reports of suspected hazards.

Despite the contributions of a variety of sources,
there is a significant lack of adequate hydrographic data.
In particular, much of the world's coastal waters must be
considered hazardous to shipping because of either
inadequacy, or age, of available survey data. Figure 1
summarizes the results of a 1975 assessment of world-wide
coastal hydrographic data (DMAHTC, 1984). The solid border
designates those areas with no survey coverage, the
diagonal hatching those with inadequate coverage, and the
vertical hatching those with adequate coverage to support
navigation. The study concluded that the data shortfall in
coastal waters in DMA areas of responsibility would equate
to over 200 years of surveying by NAVOCEANO survey ships
using existing techniques.

To more effectively manage valuable survey ship time,
DMAHTC is developing the Hydrographic Survey Requirements
System. This computer software system is intended to allow
planners to prioritize targeted areas according to the
current status of the chart production program, the amount
of traffic in the area, and the strategic importance of the
cargo being shipped through the area.

Several potentially valuable remote sensing resources
are also being developed as part of DMAHTC ' s Research and
Development program. Candidate remote sensing systems

13

%

«>!%

**m\ :a%

w3

/>■:•

///^y//^'

■/M

"W\„K

'/^"m*/.

^

f

"**».

'4

X.

X

%

^

86fc

i=

o

<

>

O

<
o

Q
<

o

<

ce

>

O

iu O

I- <

< fie

2 £

uj O
Q <->

< O

i 3

"i I

a

u

•H

f

I

3

8

u

Q)

I

W

0)
■H

0

03

C

•H
t4

14

include Synthetic Aperture Radar (SAR) , blue-green LIDAR
systems on low-flying aircraft, and satellite mul t ispectra 1
imaging systems. The last of these will be studied in
detail in this thesis.

Water surface anomalies detected in SAR imagery are
often associated with variation in bottom topography. This
relationship may offer a useful tool for DMAHTC ' s hazard
detection program.

Development of important fixed (conventional) wing
aircraft systems is also an important effort. Laser
sounding systems such as the Hydrographic Airborne Laser
Sounder (HALS) can provide high density data in relatively
little time (as compared to conventional launch surveys);
this system has already produced sounding data within the
required International Hydrographic Organization (IHO)
Standards for depths to 60m. Laser technology is also
being employed to calibrate passive mul t ispectral data
collected by conventional aircraft, as for example, on the
Multispectral Active/Passive Scanner (MAPS). (Defense
Mapping Agency Hydrographic/Topographic Center, 1984)

Ongoing research to refine algorithms for predicting
depths from remotely sensed MSS images includes development
of improved methods for measuring bottom reflectance and
water clarity, and exploration of potential uses of
multi temporal data to improve accuracy over areas with
constantly shifting bottoms.

15

Hardcopy MSS imagery has also been used to prepare
survey planning graphics in the form of zoned-depth
manuscripts (Joy, 1981). These manuscripts are compiled by
tracing areas of similar brightness, and give only a rough
estimate of relative depth; nevertheless, the manuscripts
have proven useful to planners, particularly in areas with
poor chart coverage. These developments, combined with
advances in satellite multispectral technology now being
demonstrated by the Thematic Mapper (TM) aboard Landsat D,
offer the potential for greatly increasing the
effectiveness of DMAHTC's hydrographic charting program.

C. SCOPE AND FORMAT OF THESIS

This thesis will assess the potential for utilizing
satellite multispectral-der i ved depth data in DMAHTC's
hydrographic charting program. The accuracy requirements
of this program are compared and contrasted to the
performance capabilities of current MSS depth prediction
techniques. The benefits of combining satellite
multispectral data with data from other newly developed
remote sensing systems is briefly discussed.

The first chapter defines the problem and presents
information on DMAHTC's present and projected charting
requirements and capabilities.

The second chapter contains a review of previous
research in the application of multispectral data to

16

bathymetry, and reviews the derivation of a single band
algorithm used to predict water depth from Landsat
Multispectral Scanner (MSS) data. This derivation includes
brief discussions of ocean optical properties and of
characteristics of Landsat MSS data. Published results
obtained using similar algorithms are reviewed at the end
of this chapter.

Chapter III contains the results of two case studies
conducted for the purposes of confirming previously
published results and of assessing the accuracy of MSS
derived depths under conditions approximating normal
operating procedures. Included are descriptions of the
data sets and a discussion of the processing techniques.

The analysis of results and conclusions are presented in
Chapter IV. The impact of error levels on the potential
use of MSS derived depths is the principal concern of this
discussion.

Based upon the conclusions drawn in Chapter IV, several
recommendations are presented in Chapter V. These include
both short-term implementation recommendations and
long-term research goals.

Appendix A contains a discussion of the Landsat series
of satellites, with emphasis on the MSS data
characteristics. Appendix B contains a description of the
image navigation procedures used in the case studies
described in this thesis.

17

II. WATER DEPTH ESTIMATION FROM MULT I SPECTRAL SATELLITE

IMAGERY

The estimation of water depths from measurements of
upwelled radiance is simple in concept. In areas where the
water is clear and spatially homogeneous, and where the sea
floor is composed of light colored material of high
reflectivity (compared to that of water), reflected
radiance at the sea surface will increase systematically as
water depth decreases. If both the optical properties of
the water and the reflectance of the bottom are spatially
homogeneous, then this relationship may be exploited in a
quantitative way to estimate water depth from remotely
sensed images of upwelled radiance. When this technique is
applied, using measurements from satellites above the
earth's atmosphere, it is also necessary to take into
account the radiance contributions and attenuation
associated with the earth's atmosphere.

Previous work has demonstrated at least the conceptual
feasibility of this technique using Landsat MSS images of
shallow water sites which were known to be characterized by
exceptionally clear water and bright sandy bottoms. These
results will be reviewed in this chapter, and subsequently
compared with results of two case studies analyzed as part
of this thesis project. The purpose of this analysis is to
assess the potential feasibility of applying this

18

technique, in a useful and cost-effective way, within DMA ' s
operational bathymetric survey program. Consideration will
be given to possible uses of MSS derived water depth maps
as planning tools, quality control information, and as
interpolation fields (between survey ship tracklines), as
well as to direct use in bathymetric mapping. In each
case, feasibility and utility are determined primarily by
the errors which must be attributed to the MSS derived
depth estimates.

A. THE LANDSAT MULTISPECTRAL SCANNER (MSS)

The MSS is an imaging radiometer flown as one of the
principal instruments on the Landsat series of earth
observation satellites. Each Landsat satellite is operated
in a sun synchronous orbit at approximately 900 kilometers
(km) altitude, with its ascending node occurring near 1030
local apparent time. The satellite sub-orbital track
re-visits a particular geographic location, and allows MSS
image coverage (clouds permitting) once every 18 days;
although this sampling interval makes the Landsat MSS
poorly suited for most oceanographic applications, it is
well-matched with the time scales of changes in bathymetric
features. (Storms may rework shoals extensively within a
few days, but the changes may be readily assessed by
Landsat image composites assembled over a period of several
weeks, or months).

19

The MSS images earth/atmosphere radiance patterns by
scanning a mirror over an 11.6 degree arc centered at nadir
and perpendicular to the orbital track. Upwelled radiance
is focused by the scan mirror through a telescope and
optics train into a spectrometer. Within the spectrometer,
the angular field of view of the telescope is divided into
six segment beams in the along-track direction; each of
these along-track segment beams is then directed to one of
six individual detectors in each of four wavelength bands.
During each cross-track scan of the mirror (repeated every
33 milliseconds) all 24 detectors are sampled every 9.95
microseconds to yield, for each of the four channels, six
adjacent cross-track lines of digital radiance values at
each of 3240 picture elements (pixels). A single Landsat
MSS image combines 360 successive six-line scans to form a
2340 X 3240 pixel array, with each pixel's value in each
channel representing radiance associated with a 79m X 79m
area at the earth's surface.

The MSS channels (bands) , applicable to this study,
designated four through six, each measure radiance in a 0.1
micrometer (Mm) wavelength interval centered at 0.55, 0.65,
and 0 . 75Mm respect ively; the MSS Band 7 spans a 0.4 Mm
interval centered at 0o95Mm. Of these, the wavelength
interval of MSS Band 4 spans the region of minimum
absorption by water, which makes it the most useful
channel for single-channel water depth estimation. Band 5

20

spans a region where absorption by water limits the
effective penetration depth to approximately 3m; it has
been used, together with Band 4, in a two-channel water
depth algorithm applicable to extremely shoal areas
(Polcyn, et al., 1977). In the wavelengths associated with
Bands 6 and 7 , effective optical depths are limited by
strong absorption to the top few centimeters (cm) of the
water column. (Unfortunately too, the wavelength intervals
associated with these channels are too broad to use them in
atmospheric correction algorithms of the type developed for
the Nimbus-7 Coastal Zone Color Scanner (CZCS)) (Gordon, et
al., 1983).

The characteristics of the MSS and of its digital data
are reviewed more completely in Appendix A, and in yet more
detail in the report by Taranik (1978) .

B. ALGORITHM FOR CONVERTING MSS RADIANCES TO WATER DEPTHS

The total aperture radiance Lt4 measured by the MSS in

Band 4 may be approximated by the equation

Lt4 = Lr4 + La4 + E4t4 R4 -Ejt^ R4e_2K4D + E4 t4Rh4e~2K4D t

Q Q Q (D

where total radiance Lt4 is calculated from the Band 4
digital counts as (Appendix A and Warne, 1978)

Lt4 = 8.3 DN/128, (2)

21

Lr4 is Rayleigh path radiance due to backscatter by
atmospheric gas molecules,

La4 is atmospheric path radiance due to backscatter by
aerosols ,

E4 is incident solar irradiance just below the sea surface
(E4 is estimated from solar irradiance at the top of the
atmosphere, with adjustment for time of year, single
scattering Rayleigh plus ozone attenuation through the
atmosphere, and Fresnel transmi ttance through the sea
surface. Values used are based on the similar CZCS
algorithms as contained in Gordon, et al., (1983),

R4 is the irradiance reflectance of a water column of
effectively infinite optical depth measured just beneath
the sea surface,

Q is the ratio of upwelled radiance to incident irradiance
just below the sea surface, and following Austin (1974)
it is assumed that Q = 5 (if the surface just beneath the
air-sea interface were Lambertian, then Q would equal*-),

t4 is the diffuse transmittance of the atmosphere and
ocean, which following Gordon, et al. (1983) is
approximated as:

t4 = ±-p(d) p -r§4 +ro3+/COSd

m* * (3)

22

where tr4 and t0 4 are Rayleigh and ozone optical
depths respectively (with values interpolated from those
in the same reference) , m is the refractive index of
water (assumed m = 1.33), and p (0) is the Fresnel
reflectance of the air sea interface at angle d ,
Rb4 is the reflectance of the ocean bottom, which is

assumed to be Lambertian,
K4 is the vector irradiance attenuation coefficient
(sometimes called the "diffuse attenuation coefficient"
of the water column in m~l, and
D is water depth in m.

The units of all radiance terms are mW cm~2 sr"l Mitir^,
and those of irradiance are mW cm~2 nm~*-. The subscript
"4" denotes that the value applies to the effective
monochromatic wavelength to be associated with Band 4,
which is assumed to be 0.55 Mm.

Using approximations developed for the CZCS (Gordon, et
al., 1983), the Rayleigh and aerosol path radiance terms
vary to first order as seed , where d is the viewing zenith
angle. Over an entire Landsat image, the 5.8 degree
variation about nadir in viewing angle introduces less than
0.5 percent variation in these path radiance terms and they
may be treated as constant for the scene. If it is also
assumed that the optical properties of the water (K4 and
R4) are constant over the entire ocean area covered by an

23

image, then for any deep water pixel (i.e. D is greater
than 3K4"1) the radiance L^4 may be expressed as,

Lc-=>4 = Lr + La + E4 1 4 R4
4 4 Q

(4)

which may thus be assumed to be constant over the entire
image. A measured value of L ^o 4 , obtained by averaging
over deep water pixels in a given Landsat image, may then
be substituted for the first three terms on the
right-hand-side (rhs) of (1). The fourth term on the rhs
of (1) represents the radiance reflectance at depth D from
an optically infinite water column below that depth,
assuming ocean optical properties to be homogeneous
vertically, as well as horizontally. The first four terms
on the rhs of (1) are thus seen to represent the sum of
path radiances from the atmosphere and water column of
depth D above the sea floor. The last term accounts for
radiance reflected from the sea floor itself.

Substituting (4) into (1), and assuming bottom
reflectance Rfc4 to also be constant over the imaged area,
the resulting equation may be solved for water depth D as

D = J^ In (Rb4 - R4) - l/2k4 In f(L4 - Lo~4)Q

2K4

ln["(L,

E4t4

(5

24

With the above assumptions for a particular site and image,
Equation (5) is of the form:

D = A + BX

(6

where the variable

X = In (L4 - Uo4)Q
E4t4

(7

and the coefficients A and B are related to ocean and
bottom optical properties by the equations

A = 1 In (Rfc4 - R4)

B =

2K4

-1

2K,

(8)

(9)

In use, the coefficients A and B are determined by least
squares regression for a sample of radiances extracted from
each image at control positions where depths D are known
from contemporaneous in situ measurements. Then, equations
(8) and (9) are used to calculate the values of K4 and (Rb4
- R4) to be associated with that image.

25

The principal sources of error in the above algorithm
arise from spatial variabilities in atmospheric aerosols,
ocean optical properties, and bottom reflectance.

Research with the CZCS has demonstrated that aerosol
radiance values, typically vary on scales as small at 10km
with amplitudes comparable to those of water radiance, so
that pixel-by-pixel atmospheric correction algorithms must
be used (Gordon, et al., 1983). As mentioned in the
previous section, however, the characteristics of the
Landsat MSS Bands 6 and 7 are not well suited for this
correction, and it is necessary to neglect aersol radiance
variations by assumption. As a minimum precautionary
measure, however, enhanced versions of the Band 6 and 7
images should be inspected for evidence of organized
patterns of brightness attributable to variations in aersol
density (including smoke plumes and/or brightness patterns
which appear to be spatially related to cloud patterns);
the Band 4 data should be discarded for areas where such
patterns exist.

Variations in ocean optical properties can arise from
spatial variability in concentrations of marine
phytoplankton, inorganic particulates, and/or dissolved
organic compounds ("gelbestof f e" ) (Jerlov, 1976). In
general, the MSS water depth estimation may only be applied
in relatively clear-water environments which are not fed by
a significant flux of terriginous material due to coastal

26

runoff or estuarine discharge. Therefore, waters in which
the optical properties are significantly influenced by
inorganic particles, the primary source of which is
terriginous runoff, are not candidates for this depth
estimation technique. Highly productive oceanic regions
are similarly excluded, simply because the concentrations
of phytoplankton in productive coastal waters reduce the
optical depth to a small fraction of water depth and bottom
reflections are not seen. (However, some sites may be
productive in some seasons, but relatively barren in
others, so that in certain months water clarity may permit
estimation of bathymetry from MSS images) . Since the
major sources of dissolved organic compounds in the ocean
are humic compounds contributed by terrestrial runoff and
decayed phytoplankton materials in highly productive
waters, variability of "gelbestof f e" concentration would be
a factor only in situations which would be otherwise
excluded from consideration here on the basis of more
easily detected contributors to ocean optical properties.

The reflectance of the sea bed is dependent upon the
composition, roughness, and to some extent, upon the slope
of the bottom. Of these three characteristics, a change in
composition (which here refers to both lithographic and
vegetative characteristics) is the most significant factor
in estimating bathymetry from reflected radiance. The
water depth algorithms under consideration attribute change

27

in signal value to variation in the depth of a bottom of
constant reflectance. However, if the bottom reflectance
was assumed to be that of a very bright, sandy bottom, and
a pixel was observed an area of dense bottom vegetation,
the predicted depth would be much greater than the actual
depth. The reverse case is also possible, although
conservative depth estimates are preferable as they pose no
actual danger to ships. Errors in depth estimation
resulting from assuming a constant seabed composition are
possible at all depths which can be sensed and require a
conservative estimate of Rj-,4 to bias errors towards
underestimates. This "safe bias" procedure will be
discussed in conjunction with error analyses in Chapter II,
IV, and V.

The effect of texture changes, independent of
composition variation, should be significant only in
shallow waters (with depths less than 5m). This is because
texture variation affects reflection of direct radiation,
and downwelling light becomes increasingly diffuse with
depth. The effects of texture changes upon reflected
radiance are further reduced since the seabed itself acts
as a diffuse reflector. For these reasons, the effect of
slope changes should also be significant only in shallow
waters with steep banks (Warne, 1978).

The residual standard deviation of the regression
analysis of the control sample used to determine

28

coefficients A and B provides a crude measure of the
combined effects of spatial variability of bottom
reflectance and water optical properties. In situ optical
observations, made in conjunction with measurement of water
depths at control positions, can (at least in principal) be
used to separate these contributions. It is particularly
important to detect the presence of patchy variations in
phytoplankton concentrations and/or bottom vegetation,
since either occurrence can lead to potentially serious
overestimates of water depth. Techniques for routinely
measuring optical properties during in situ bathymetric
surveys are briefly discussed in Chapter V.

Equation (1) neglects reflected skylight, and assumes
that sun glitter is negligible. Since this water depth
algorithm is based on relative variations in upwelled
radiance over a region spanning both deep (optically
infinite) and shallow water, and since it is reasonable to
assume that skylight is constant over a nearly cloud-free
Landsat image, the neglect of reflected skylight will, at
worst, introduce a bias error which will be absorbed in the
model's empirical B coefficient (Equation 5). Errors due
to neglect of sun glitter can be more serious at low
latitudes in summer months. With solar zenith angles less
than 30 degrees, sun glitter can be significant even for
near nadir viewing angles (Cox and Munk, 1954); this error
source can be especially serious in areas where small scale

29

variations in surface winds (e.g., near coasts and islands)
introduce variations in glitter radiance. Therefore, this
algorithm should be used with extreme caution if applied to
an image associated with a solar elevation much in excess
of 55 degrees.

C. PREVIOUS RESULTS

In the late 1960's Brown et al. (1971) began developing
techniques for using passive multispectral scanner data for
bathymetry. By 1973, the basic principles of deriving
depths from single band data had been demonstrated using
both aircraft and MSS data (Brown, et al . , 1971).
Development efforts were then extended to two-channel
techniques which were designed to reduce the sensitivity of
the water depth algorithm to variability in bottom
reflectance and optical properties of the water column. In
tests conducted over the Little Bahama Bank, with low gain
Landsat MSS data, single-band techniques were used to
predict depths to 9m, and the dual-band method was used to
predict depths to 3m (Polcyn and Lyzenga, 1975).

In 1975, the National Aeronautics and Space
Administration (NASA) and the Cousteau Society jointly
sponsored an ocean bathymetry experiment designed to
determine 1) the maximum depth measurable using high gain
MSS data from Landsat 1 and 2, and 2) the potential
accuracy improvement of the MSS depth estimates when

30

supported by in situ measurements of relevant ocean optical
properties. Tests were conducted in two areas of different
optical properties. The first area was located on the
northern edge of the Great Bahama Bank, west of the Berry
Islands. This area was chosen for its relatively high
bottom reflectance (determined to be 26 percent) , and water
clarity (K4=0 . 058m"1 j . The second area was located off
the east coast of Florida, near Hollywood. For this area,
the bottom reflectance and irradiance attenuation
coefficient, K4, were determined to be 20 percent and
O.llm"^ respectively. The accuracy of the MSS derived
depths was determined through comparisons with fathometer
readings taken at the time of satellite imaging (Polcyn,
1976) .

At the Berry Islands Test Area, MSS depth measurements
were determined to depths as great as 22m with an rms error
of 10 percent, and bottom reflected radiance patterns were
detected in the MSS images from depths as great as 40
meters. During this study, depth maps were prepared from
the Landsat-derived data for that portion of the Berry
Island Test Area covered by N.O. Chart 26320. Several
different processing techniques were tested in the
preparation of the depth maps, including smoothing to
reduce the effects of varying responses among each of the
satellite's six sensors, and in turn, enhancements to
reduce the coarsening effect of smoothing (Polycyn, 1976) .

31

In 1911, Hasell reported on the development of the M8
active/passive scanner system. This system combined a
mul tispectral scanner with a pulsed laser system (Hasell,
1977) and was designed to be carried aboard conventional
aircraft. The bathymetric applications of the M8 system
were tested over several sites in the Bahamas in August
1978, under a contract with DMA. The depth data from the
active laser system was used as ground truth to compute a
relationship between bottom depth and passive scanner
signal value. This formulation was then used to predict
depths, given only scanner values. For the North Cat Cay
test site , depths under 15m were predicted with an rms
error of 0.72m, and for the Bimini test site, with an rms
of 0.923m (Lyzenga, 1981).

The potential usefulness of depth maps prepared from
high gain MSS data in pre-survey reconnaissance work was
demonstrated by the Environmental Research Institute of
Michigan (ERIM) , in 1977 under a contract to DMA (Polcyn,
et al., 1977). Seven depth-zone maps were produced for
portions of the Little and Great Bahama Banks, using water
depth algorithms developed previously. The maps were then
compared with the existing chart coverage, and the
satellite maps revealed several areas requiring
re-surveying.

An independent study by Warne (1978) addressed practical
applications of Landsat imagery to hydrographic mapping,

32

with specific emphasis on applications of the technique in
waters near Australia. Warne developed an interactive
computer software system called Hydrographic Mapping System
(HYDMAP) , which allows an interpreter to interactively
segment an MSS image into areas of consistent optical
characteristics. Using HYDMAP, and applying additional
radiometric corrections to each Landsat image, Warne
prepared zoned depth maps for two test areas in the waters
between Cape York, Australia, and Papua, New Guinea. The
depth prediction techniques used were found to be accurate
to within lm rms to a depth of 16m.

Research on this topic meanwhile continued in the
United States. To further evaluate the accuracy of
single-channel depth prediction techniques, Lyzenga and
Polcyn (1979) compared depths derived from Landsat MSS to
leadline and fathometer sounding data at 10 stations in the
Bahamas Phot oba thyme tr ic Calibration Area. The water
attenuation coefficient used was the same as that measured
for the Great Isaac station during the NASA/Cousteau
Photobathymetric Survey Study of 1975-1976 (Lyzenga and
Polcyn, 1979) . Bottom depths were predicted using two sets
of bottom reflectance values. The first set of
calculations used the value of 0.22; this value had been
used previously in processing the same scene. A second set
of calculations were based upon bottom reflectance values
for each station which were derived from underwater

33

photographs of the bottom and test panels of known
reflectance (Lyzenga and Polcyn, 1979) .

Tables 1 and 2 present the results obtained using each
set of bottom reflectance values. The following
information is presented for each station: measured depth,
MSS-derived depth (predicted depth) , the bottom reflectance
value used to predict depth (Rfc4) , and the difference
between measured and. predicted depth. The standard error
of prediction was 2.6m using the assumed value Rb4=0.22 and
1.9m using individually measured values of RD4«

TABLE 1

ERIM S

tudy-Resu

Its

Using Sing

le Bottom

Reflectance Value

Station

Measured

Predic

ted

Dm-DP* Rb4

Depth

(Dm)

Depth

(Dp)

C-5

9.8m

7.7m

+2.1 0.22

D-7

9.1m

•

9.1m

0.0 0.22

P-ll

9.8m

11.5m

-1.7 0.22

G-12

10.4m

11.5m

-1.1 0.22

H-13

4.9m

3.3m

+1.6 0.22

1-14

6.7m

8.3m

-1.6 0.22

J-16

12.5m

16.5m

-4.0 0.22

A-17

6.1m

6. 7m

-0.6 0.22

B-18

10.7m

15. 3m

-4.6 0.22

D-21

6. 1m

9.1m

.

-3.0 0.22

Standard Error of Prediction 2.6m*
Mean Bias = -1.29*
Source of Tables: (Lyzenga and Polycn, 1979)
*Derived Values

34

TABLE 2

ERIM

Stu

dy-

-Result:

s Using

Indivii

dual

Bottom Reflectance Values

Station

Measur

ed

Predic

:ted

Dm-DP* Rb4

Depth

(Dm)

Depth

(Dp)

C-5

9.8m

7.1m

+2.7 0.20

D-7

9.1m

9.7m

-0.7 0.24

F-ll

9.8m

11.8m

-2.0 0.23

G-12

10.4m

10.5m

-0.1 0.19

H-13

4.9m

3.9m

+1.0 0.24

1-14

6. 7m

3.0m

+3.7 0.10

J-16

12.5m

11.2m

+1.2 0.10

A-17

6. lm

7. 0m

-0.9 0.23

B-18

10.7m

12.7m

-2.0 0.15

D-21

6. lm

7.0m

-0.9 0-16

Standard Error of Prediction 1.9m*
Mean Bias = +0.2*
Source of Tables: (Lyzenga and Polycn, 1979)
♦Derived Values

Assuming constant water clarity from site to site, as
Lyzenga does in this investigation, the relationship
between measured depth and predicted depth reveals the
effect of varying the bottom reflectance value. The assumed
constant value of 0,22 yielded deeper depths than were
actually measured at all but three stations; this is
reflected in the negative value (-1.29m) of the mean bias.
This result indicates that the true bottom reflectance at
the majority of the stations was less than 0.22., an
inference confirmed by inspection of actual reflectances in
Table 2. When the lower, measured bottom reflectance
values were used at those stations with negative residuals
(G-12, B-18, and D-21, for example) predicted depths were
much closer to the measured depths (Table 2) . Similarly,

35

for those stations at which the inferred value gave depths
which were too shallow (C-5 and H-13 in Table 1), the
higher measured values yielded better results (Table 2).
Accordingly, the standard error of prediction improved
(from 2.6m to 1.9m). The mean bias also became
significantly smaller and shifted to a positive value,
indicating more conservative predictions. When the
measured bottom reflectance value was greater than 0.22,
for those stations at which the inferred depth values were
too deep, the result was even greater residuals, as shown
by stations F-ll and A-17. Similar reasoning can be used
to explain the results obtained at each of the other
stations .

The tendency, noted above, for estimated depths to be
dangerously biased towards overestimates, points to the
need to independently estimate the spatial variabilities in
ocean optical properties and bottom reflectance using
concurrent ground truth data. Further, it will be
desireable to use some minimum probable value for bottom
reflectance RD47 this procedure will bias MSS depth values
towards underestimates, and reduce the probability of depth
overestimates greater than some critical value (e.g. 0.3m)
to a level to be determined by future research. This "safe
bias" concept is discussed in more detail in Chapters IV
and V.

36

For five stations there was a loss of accuracy using the
measured values; the measurement of bottom reflectance was
unsatisfactory. Lyzenga, in fact, proposes that the bottom
photographs used to measure bottom reflectance may have
favored the bottom vegetation more than the sand, and/or
that they reduced the contrast between very dark, and very
bright areas (Lyzenga and Polcyn-, 1979) . In either case,
the bottom reflectance was the average value for the
instantaneous field of view of the Landsat sensor.

A second possible explanation for the reduced accuracy
at these five stations is that the initial assumption of
constant water clarity was incorrect. Lyzenga states that
conditions were unsuitable for obtaining in situ
measurements during the time of the experiment, so that no
comparison set was available* However, this issue is
discussed further in section III.E, where Lyzenga and
Polcyn 's results will be compared with those obtained for
the same Landsat scene in the present investigation. These
two experiments will then form the basis for the discussion
of error impact analysis and potential uses for satellite
multispectral imagery in bathymetric mapping (Chapters IV
and V) .

D. SYNOPSIS

The principles of producing water depth from single-band
MSS data are sufficiently understood to begin to assess the

37

potential role of this tool in bathymetr ic survey
operations. Research has shown that the usefulness of
these procedures is limited to clear, homogeneous waters
with bright bottoms, precluding their use in areas of high
organic productivity or in coastal waters receiving
significant terrigenous runoff. Likewise, the principal
error contributors have been identified as variability in
atmospheric aerosols, ocean optical properties, and bottom
reflectance. Research efforts have also resulted in the
preparation of potentially useful prototype products.

The first step in defining the usefulness of MSS-derived
bathymetric products is to quantify expected error levels
associated with the predicted depths. Two case studies
(designed to simulate operational procedures) will be used,
in conjunction with those results reported by Lyzenga and
Polcyn (1979), to derive values for expected error levels.
These values will then be used to assess the cost benefits
of combining multispectral data with control tracklines to
obtain nearshore hydrographic data.

38

Ill . SELECTED CASE STUDIES

A. INTRODUCTION

The previous experiments reviewed in Chapter II have
indicated the potential usefulness of Landsat MSS-derived
depths to the hydrographic community. To more clearly
define the potential role of this tool, two case studies
were conducted. The studies were designed to reexamine
published results under conditions approximating
operational procedures, and to assess the applicability of
these procedures under various ocean optical conditions.
Much of the discussion in Chapter IV, on the impact of
error sources, and on the cost benefit aspects of this use
of Landsat MSS data, is drawn from the results of this
investigation.

B. SITE CHARACTERISTICS

Several geographic areas were initially identified as
being suitable for this investigation. These were selected
for their relatively high, and reasonably homogenous,
bottom reflectances, and for their variety in water
clarity. Final site selection was based upon the
availability of both high-gain Landsat tapes and
contemporaneous survey data. Four sites were chosen from
two test areas satisfying the selection criteria.

39

Test Area 1 is located on the northwest-western margin
of the Great Bahama Bank. Two sites were selected from
this test area. Site 1 is located in the waters south and
west of the island chain formed by South Cat Cay, North Cat
Cay, and Gun Cay (Figure 2). The site measures
approximately 5.5 nautical miles southeast to northwest and
approximately 1.75 nautical miles northeast to southwest.
Site 2 is approximately 15 nautical miles from Site 1,
located off the northwest coastline of North Bimini (Figure
3). This site measures nearly 4.5 nautical miles north to
south and averages 2 nautical miles, on the average, east
to west. These correspond to portions of areas 1 and 3,
respectively, of the Bahama Photobathymetr ic Calibration
Survey of 1980.

Test area 2 is located on the west Florida shelf between
28° 10' N and 27°30'N. As with the Bahamas Test Area, data
was collected from two sites within Test Area 2; these are
labelled Site 3 and Site 4 (Figure 4). The location and
seaward extent of these sites was fixed by the boundaries
of the survey from which the set of control depths were
taken. Site 3 extends approximately 15 nautical miles off
the Florida coast at Tarpon Springs and measures almost 6
nautical miles north to south. The orientation and
evenness of the depth contours in Figure 4 reflect the
steady increase of depth with seaward distance. For this
reason data spanning a suitable range of depths was

40

19'

I
17'

79" 15

-34

-32

25°
"30-

SITE 1

NORTH CAT CAY

SOUTH CAT CAY

* Approximate representation
of depth curves only

34-

32-

25-

30'

19

17'

79"15

Figure 2. Case Study 1 - Site 1

41

18'

79°16

* Approximate representation
of depth curves only

-48'

SITE 2

-46'

-44

25°
"42'

q NORTH RK

79°16

48 -

46 -

144 -

25'
42'"

Figure 3. Case Study 1 - Site 2
42

-10'

28°
"00'

-50

-40

27°
"30'

\
J

"^

o

33°00-

"I

r

~TT'

LINE A \

SITE y"3

LINE B

/

\

!

J

S

/

k

\

1

5

o

V

SITE 4

\

LINE C

A.

TARPON
SPRINGS

10'-

50-

^

Approximate representation
of depth curves only

ANNA MARIA
ISLAND

27°
30'

33"00

Figure 4. Case Study 2 - Sites 3 and 4

43

extracted by sampling along transects perpendicular to the
shoreline; in Site 3, these are labelled Lines A and B.

Site 4 is located approximately 35 nautical miles south
of Site 3, and extends almost 30 nautical miles offshore.
The difference in seaward extent between Sites 3 and 4
reflects the requirements of the original survey from which
control depths were drawn.

C. DATA DESCRIPTION
1 . Landsat Images

The Landsat images used in this investigation
received the standard NASA bulk processing applied to all
pre-1978 MSS data (Appendix A). The data was released in a
"band-interleaved by pixel-pair format" (ERTS-f ormat ) , on
9-track, 1600 bpi computer compatible tapes (CCTs). The
two MSS images analysed in this project were obtained from
DMAHTC, NASA, and Goddard Space Flight Center. The
characteristics of each image are summarized in Table 3.

44

TABLE 3

Identification of Landsat-1 MSS Imagery and Key

Radiation Parameters for Case Study Areas 1 and 2

Test Area 1 (Great Bahamas) 2 (West Florida

Shelf)

Recording Mode High Gain High Gain

Date , 24 December 1975 13 March 1976

Time 1443Z 1435Z

Solar Elevation 2 8 4 8
Incident Solar

Irradiance

(E4) (mWcm~2 m-1) 85.759 139.55
Deep Water

Radiance (L /)

(mWcm~2sr~1 nrfj 2.853 4.766

Cloud Cover less than 10% less than 10%

Scene Number 5249-1443500 85360-1453500

2. Water Depth Data

Water depths used to estimate the precision of
depths calculated using Equation (2) -(5) and Landsat MSS
radiances were taken from recent NOS hydrographic survey
sheets and from NOS charts. In the Florida Test Area, both
sources of comparative depth data were used (Table 4).

TABLE 4
Sources Used for Control Depths: West Florida

Shelf Test Area

Source Date Scale

NOS Survey Sheet H-9509 4/75 - 8/75 1:20,000

H-9510 4/75 - 8/75 1:20,000

H-9338 4/75 - 8/75 1:20,000

NOS Chart 1140 19th Ed. 8/15/81 1:40,000

11412 25th Ed. 5/17/80 1:80,000

11414 27th Ed. 3/27/82 1:40,000

11424 13th Ed. 4/25/81 1:80,000

45

Since the survey data are more accurate, the NOS charts
were used only where the survey data did not adequately
cover the Test Area and in cases where densely clustered
charted depths showed no depth variation over the sampling
area. All depth data used in the Great Bahama Bank Test
Area was extracted from survey data collected by the R/V
Adam's Folly in 1980, in conjunction with the Bahamas
Photobathymetry Study being conducted by DMAHTC . This Test
Area has been frequently surveyed and the data showed no
significant variations in bottom topography over the period
1976-1980 (James Hammack, DMAHTC, Personal 'Communication,
1981). The control depths over the West Florida Shelf were
taken primarily from survey data collected less than one
year earlier than the corresponding Landsat image. Since
there were no effects from major storms in the intervening
months, it is assumed that the Landsat images and the
comparison depth samples are effectively contemporaneous.

D. METHODS OF ANALYSIS

The relationships between water depth, D, and Landsat
counts are expressed in Equations (4), (5), and (6). For
each test area, samples were prepared in which computed
values of the log reflectance anomaly, X, as defined in
(7), were paired with survey depths.

A navigation algorithm was developed to identify the
pixels corresponding to the geographic position of each

46

sample depth. The procedure was tested using landmarks
identifiable on the Landsat scenes, yielding a circular
error (at the 90 percent confidence level) of 113.3m for
the Great Bahama Bank scene, and 143.5m for the West
Florida Shelf scene (Appendix B) .

In all cases, both X and D values represent the average
over a ground area roughly equivalent to the appropriate
navigational circular error. Therefore, each MSS4 sample
value represents the average of a 3x3 pixel matrix for the
Bahamas scene, and of a 4x4 pixel matrix for the Florida
scene.

The average control depth values used in the Bahamas
test area were computed by first plotting track lines and
speeds of the survey vessel from the positional and timing
information on the data tape, and then determining the
number of sounding events covering an area 216m square
(equivalent to the navigational circular error). Because
ship speed varied from track line to track line, and
because the density of track lines also varied, these
computations were made separately for each track line used.

Average depth values from the West Florida Shelf Test
Area were computed by averaging each plotted depth (from
either a sounding sheet or a chart) within the area
equivalent to a 187m square area (corresponding to the
circular error) .

47

Survey depth values for each test area were corrected to
match the tidal stage at the time of Landsat MSS
observation. Following Polcyn et al. (1977), a correction
factor of +0.61m was applied to all Bahamas depths. Tidal
corrections for the West Florida Shelf depths were computed
from data supplied by the Tides Branch of the National
Ocean Survey (NOS), Rockville, Maryland. Predicted tidal
height for Site 3 at the time of the Landsat observation
was calculated to be 0.55m above Mean Low Water (MLW)
(predicted form the Clearwater Tidal Station); predicted
tidal height for Site 4 was calculated to be 0.13m above
MLW (predicted from the St. Petersburg Tidal Station) . As
with the Great Bahama Bank data set, all West Florida Shelf
tidal corrections were applied to the surveyed depths, to
bring these into agreement with the Landsat-der i ved depths.

For each sample, a scatter plot of X versus D was
prepared and inspected for possible outliers. Each
suspected outlier was tested using the statistic and Table
of Percentage Points for the Studentized Maximum Absolute
Deviate prepared by Halperin (1955). Each data point which
failed the initial outlier test was re-evaluated by
comparing it's MSS4 and depth sampling statistics to those
computed for the entire sample set. If the local variance
of the outlier cell was less than the average local
variance for the overall data set, the point was discarded.
On the other hand, if the local variance for either depth

48

or MSS4 count exceeded the average of the entire sample
set, local outliers were discarded and a new depth and/or
count average was computed. A new outlier test was then
performed with the revised values, and any points again
failing the test were discarded.

The sample for each test site was divided into a control
sample and a test sample. For each site (excluding Site 2,
which had too few points) a simple linear regression was
performed to compute the coefficients A and B of Equation
(5), using data from the test sample. From the values of A
and B, local values for K4 and R^^ were calculated from
Equations (8) and (9). These values were compared, for
plausibility, to published representative values for the
region (Jerlov, 1976). Equation (5) was then used to
compute corresponding depths D for each X (Equation (7))
value in the control sample.

An additional linear regression analysis was performed
for each of the two test areas using the combined test
samples from each of the two included sites. The predicted
depths obtained from this set of coefficients were compared
to the former set to assess the regional dependence of the
technique.

E. RESULTS

The points selected for each test sample are depicted
graphically in the scatter plots of Figures 5-8; the

49

o

CO
C\J

C\J

c\j
c\j

o

CO

in

CM

O

CM

in

o .— i

2

><

Q ®

CJ CO

rd +j

4J en

H

M-i —I

OS O

W

I

n-i O
0 ^

a,

in
(D

•H

(S U 3 1 3 W) H 1 d 3 Q

50

o

CO

00

lO

• CD

•

M H

CM

naly (X) v
Test Samp

^

a -a

5 G

CM

eflectance
ite 2 Caubi

CM

Qi 01

CM

of Log
1 and

UJ

0

■y -t-1

=>

0 -H

.H 01

o

_1

PL|

y

CM

<

0

• 4-1

>

CD

X

•H

fa

00

CM

O
CM

LT)

O i-H

(S H 3 1 3 W) H 1 d 3 Q

51

« «

C\J

o

ro

s

CO

•

,^— N

CM

arameter (X
e

lO

CU rH

•

CM

eflectance
3 Test Samp

«3-

OS

CM

of Log
for Sib

LU

tj

ID

H

0,

CM

_1

CM

<

>

1

X

•

r*-

0

•H
En

O

CM

CM

o

CM

Lf>

IT)

CO

O «-H

(SU313W) Hld3Q

52

CO

10

«a-

w

>

m

nomaly (X)
e

CM

< i— i

•

CO

Reflectance
4 Test Samp

o

(D

en

of Log
for Sit

LU

t5

=>

H

CO

_l

C\J

<

>

1

CO

(D

X

.1

Uj

C\J

CM

O

CM

O CM

(SH313W) H 1 d 3 Q

53

presence of a linear correlation is strongly suggested in
each figure.

The following statistics were computed for each sample
set: 1) the squared correlation coefficient R^, 2) the
standard deviation of the residuals Sq in m, and 3) the
standard error of prediction Sp in m. These values are
presented in Table 5, together with the linear regression
coefficients (A and B) and the derived optical parameters
(K4 and Rb4) •

TABLE 5

Results of Analysis for Test Areas 1 and 2

SITE (S)

K4,

(m-1)

1

.041

1 & 2*

.052

3

.045

4

.043

3 & 4*

.043

Rb4

B

(m)

0.204 -24.107 -14.891 0.88 1.90

0.258 -16.482 -11.869 0.79 1.26

0.237 -20.642 -11.202 0.79 0.56

0.236 -25.228 -11.548 0.83 1.08

0.234 -29.993 -12.879 0.74 1.15

Sp
(m)

3.86
2.27
1.65
1.92
1.84

(*Combined)

The depth predictions in Test Area 2 (Sites 1 and 2)
were consistently more accurate than in Test Area 1 (Sites
3 and 4). The largest Sp computed for Test Area 2, 1.92m
(Site 4), was 15 percent lower than the 2.27m Sp computed
for 3.86m Sp computed for Site 1. The Sp for each of the
three analysis performed in Test Area 2 were likewise
consistently lower than those computed for Test Area 1.

The computed statistics also indicate the effect of
increasing sample size in each test area. In Test Area 1,
the Sp improved from 3.68m to 2.27m by combining the sample

54

from Sites 1 and 2; similarly, SD was lowered from 1.90m to
1.26m by pooling the data from the two sites. In Test Area
2, no trend emerges as a result of combining Sites 3 and 4.
The resulting SD was lower than that computed for Site 4,
but was higher than that computed for Site 3. The R2 and
SD values computed for the combined Sites 3 and 4 (0.74 and
1.15m) were actually degraded from the values computed for
each of these statistics for Sites 3 and 4 separately.

There were also significant differences in the
homogenity of the two test areas. The optical
characteristics (derived from the coefficients A and B) for
Sites 3 and 4 are more internally consistent than are those
for Sites 1 and 2. The values for K4 range from .041 to
.052 in Test Area 1, but only vary from .043 to .045 in
Test Area 2. Similarly, Rj-,4 values range from 0.204 to
0.250 in Test Area 1 and from 0.234 to 0.237 in Test Area
2.

The analysis of results presented in Charpter IV will
address the relative accuracies achieved in the two test
areas, the homogeneity of the optical factors, and the
effect of varying sample size in each test area. Assuming
these factors are related, the site dependence of this
technique and the extent of the error analysis necessary
for areas of varying homogeneity is indicated. These
issues are discussed in the Cost Benefit Analysis presented
in Chapter IV.

55

IV. DISCUSSION

A. ANALYSIS OF RESULTS

The linear regression analyses performed on the Great
Bahama Bank and the West Florida Shelf samples indicate
significant differences between the two test areas. The
results obtained from Test Area 1 are characterized by
internally inconsistent values for the optical
parameters, K4 and Rb4/ relatively inaccurate depth
prediction (compared to Test Area 2), and some improvement
in accuracy with increased sample size. Test Area 2
yielded consistent optical values, relatively accurate
depth prediction, and no increase or decrease in accuracy
with varying sample size. The interpretation of these kinds
of differences is an important step in determining the
conrol required to achieve desired accuracies.

The magnitude of the differences in the K4 and Rb4
values derived for the two analyses performed in the Great
Bahama Bank Test Area indicates the need to acquire more
representative test samples. The strong tendency of the
model to over-estimate depths in shallow waters (less than
11m) in both analyses, emphasizes the need for improving
the accuracy level. Lyzenga and Polycn (1979) obtained
similar results, as indicated by a mean bias of -1.29m,
using a single bottom reflectance value for several
stations, further emphasizing the dangers of not accounting

56

for inhomogeneities . For this test area, the accuracy of
the predictions might be improved by either acquiring a
larger combined sample, or by acquiring larger local
samples from each site and performing separate linear
regressions. Lyzenga and Ploycn's results for the same
area support the latter alternative. The wide range in the
bottom reflectance values measured (0.10 to 0.24) is a
good indication that no single set of coefficients will
ensure accurate depth prediction for the entire Bahamas
Photobathymetr ic Test Range. This result suggests: a) that
spatial trends in optical properties be determined using
concurrent ground truth data, and b) that a procedure for
estimating a "lowest probable bottom reflectance" which
will bias MSS depth errors towards relatively safe
underestimates. These suggestions will be developed in
more detail in the next section and in Chapter V.

The similarity in the optical parameters derived for
each of the three West Florida Shelf sample sets indicates
highly homogeneous conditions over the entire test area.
These optical conditions, and the smoothly varying bottom
topography of the West Florida Shelf, combined to yield
greater accuracy than was achieved in the Great Bahama
Test Area. The accuracy achieved by combining samples from
the two sites was higher than that achieved with the Site 4
sample but less than that with the Site 3 sample. This
result was anticipated since, in a nearly homogeneous

57

environment, a relatively small sample may be just as
representative of the entire population as a larger sample
would be. Likewise it is obvious that, in this instance, a
sample collected from a single site is potentially as good
a predictor of depths throughout the test area as a sample
collected from several sites.

The differences between the results obtained from the
two test areas are seen to be a result of the differences
in the degree of homogeneity in each area. In Section IV. B
these results will, in turn, be used to assess the impact
of various optical conditions upon the usefulness of single
band prediction techniques.

B. ERROR IMPACT ANALYSIS
1 . Introduction

The error associated with water depths derived from
MSS imagery is the primary factor governing the scope of
their use in a bathymetric charting program. The
requirements for depth data intended for use in nautical
chart compilation are more stringent than for those used in
pre-survey evaluation. It is also apparent, from the
results of the two case studies performed for this thesis
project, and from the results of Lyzenga and Polycn (1979),
that the quantity of control data required to achieve a set
level of accuracy will vary from area to area.

58

2. Sources of Error in Depth Estimation

Variations in bottom reflectance, water clarity,
and/or any other optical values assumed constant, will
contribute errors in depth prediction. In the Bahamas Test
Area, the variation in bottom reflection was identifiable
as a principle error contributor. This case is probably
typical of coral reef environments, which exhibit marked
variability in bottom vegetation. Usually vegetation acts
to reduce bottom reflectance, especially relative to that
of the bright sandy bottom typical of the Bahamas.
Therefore, if the average bottom reflectance to be used in
an area has been measured over a relatively vegetation-free
bottom, increased vegetation at any location will result in
overestimated depths. Underestimation of depths occurs if
the "average" bottom reflectance is measured over an area
with more than the normal amount of vegetation.

The obviously dangerous risks associated with
overestimated depths suggests that a procedure be developed
to determine a "lowest probable bottom reflectance", which
will bias errors in MSS depths towards less dangerous
underestimates. Future research should be initiated to
standardize such a "safe bias" procedure, and to quantify
the associated probability of overestimated depths being
greater than 0.3m.

Although the Bahamas test results did not so
indicate, a second possible error source in areas of active

59

reef building is the variation in water clarity caused by
the presence of plankton. Planktonic activity increases
the attenuation of light, thereby reducing the bottom
return; this results in overest imat ion at all depths.

The results summarized in Table 5 confirm the
assumption that the bottom reflectance in the Florida Test
Area is more spatially uniform than that in the Bahamas. It
was expected however, that there might be some noticeable
difference in water clarity between Sites 3 and 4. Site 4
is located nearer the mouth of the Tampa Bay, and it would
not have been surprising if either or both bottom
reflectance and water attenuation would have changed
systematically between open coast and estuarine
environments. However, systematic variations were not
evident in either K4 or Rb4»

C. IMPACT OF ERRORS ON POTENTIAL USES

1 . Chart Evaluation Using MSS Derived Bathymetry

Zoned-depth maps prepared from multispectral
imagery may be used to evaluate existing charts. Initially,
the boundaries between depth zones could be compared to
bathymetric contours to detect major topographic changes
over time. Because patterns, rather than isolated
soundings are being compared, the absolute vertical
accuracy requirements are less stringent than for direct
mapping using MSS alone.

60

Although it would be desirable to compute depths
from local groundtruth, this will usually be impractical
since the survey data being evaluated is questionable, and
K4 and Rb4 will generally not be known to any useful degree
of accuracy.

However, even visual comparisons of zones of
relative brightness on MSS imagery with depth contours have
proven to be useful. Such comparisons have resulted in
detection of at least one uncharted marine hazard which was
later verified (Hammack, 1977). At the very least, the
zoned depth map can be used to produce a relative
evaluation of chart coverage, for better long range survey
planning .

2 . Use of MSS-Derived Depths as a Pre-Survey Planning

Tool

The purpose of pre-survey planning is to determine
the most cost effective procedure for accurately gathering
adequate hydrographic data for a prospective site. When the
data available from a particular survey site are unreliable
or incomplete, MSS coverage can provide a valuable planning
tool. DMAHTC has prepared several relative depths maps for
use by NAVOCEANO in the planning stage (Joy, 1981) .
Boundaries were defined to correspond to depth contours
used on nautical charts. Typically 0.5m, 10m, and 20m
isobaths, and if high tide and low tide images are both
available an "uncovers" curve, were portrayed. Comparison

61

of corresponding isobaths plotted on each chart type
indicated the degree of change in bottom topography since
the last survey. Previously undetected shoals were
indicated on the MSS-depth map, and some charted dangers
were confirmed; both occurrences represent cases requiring
special attention during the next survey. Because
undetected shoals represent danger to survey ships, as
well as to normal shipping, detection of doubtful depths
alone may justify the expense of preparing depth maps from
MSS imagery.

The relative depth curves depicted on the initial
maps prepared for NAVOCEANO were based almost entirely on
variations in brightness (photointerpretation techniques
were used when possible to classify bottom types). Although
no accuracy level was attached to these products they were
found to be useful planning tools.

Although depths predicted from the algorithm
described in this study are potentially more reliable than
"raw" MSS4 counts, the cost of obtaining the necessary
ground truth does not appear to be justified for the
purpose of pre-survey planning.

For many unsurveyed, or insufficiently surveyed
areas, estimates of water clarity and bottom type can be
made with sufficient accuracy to make use of MSS depth data
in prioritizing and planning surveys. The inability to
define a level of accuracy does not preclude these

62

applications entirely. The potential benefits include more
informed risk analysis in deciding where charts most need
to be updated because of possible dangers, and more
informed pre-survey planning to ensure that possible
dangers are investigated and developed thoroughly. These
benefits easily translate into safer waters.

3 . Use of MSS Imagery as a Direct Source of Bathymetric

Data

Depth information appearing on nautical charts must
be both accurate and reliable. The errors associated with
depth prediction in this and previous studies exceed the
errors allowed for conventional survey data (0.3m for
depths between 0 and 20m, and 1.0m for depths between 20
and 100m) (Umbach, 1976). Because these studies were
conducted in locations with above average bottom
reflectance and/or water clarity, lesser accuracies may be
expected or achievable in less ideal sites.

Also the risk of overpredicting depths, especially
in very shallow waters, is unacceptably high. All results
of the present and previous work indicate strongly that MSS
estimates of depth are too unreliable to be used as a
direct independent basis for bathymetric maps. Practical
charting applications will be restricted to techniques
using MSS imagery together with other, independent sources
of bathymetric data.

63

4 . Use of MSS-Derived Depth Data as an Interpolation

Tool

The accuracies indicated by this and previous
studies (Lyzenga, 1981) suggest the possibility of using
MSS-derived depths for interpolating between survey track
lines in some areas. The potential usefulness of the
technique in a particular candidate survey site can be
evaluated through pre-survey analysis of MSS imagery. This
survey planning application of MSS images is discussed
above and in Chapter V.

Conventional survey track lines are planned at
specified intervals to insure complete coverage and to
optimize use of ship time. By supplementing shipboard
survey data with a matrix of accurate and reliable depths
produced from MSS imagery, it would be possible to widen
track lines while still meeting data acquisition
requirements. Optical measurements taken at specified
intervals during the survey would be used to compute the
coefficients required for predicting depths with MSS data.
Using commercially available microcomputers and submersible
optical instruments, a towed optical measurement package
can be configured to obtain continuous along-track
measurements with only minimal operator attention during
routine surveys. If K4 and R4 are measured in situ along
each survey trackline, it will be possible to constrain the
regression fit to account for spatial trends in ocean

64

optical properties and bottom reflectance. With such
spatially adjusted constraints, previous work (Warne, 1978)
has demonstrated that rms accuracies of less than lm can be
expected (additional research must be done to determine
whether accuracies approaching 0.3m are feasible).
Additionally, use of spatially adjusted minimum bottom
reflectance can reasonably ensure that errors in the depth
interpolation field are safely biased underestimates.

The accuracy level of the interpolated depths would
be calculated by comparing the predicted depths along each
track line with the reduced sounding data taken along
randomly interspersed intermediate survey tracklines. This
method of obtaining a reserve sample, and determining rms
errors in interpolated soundings, can be easily extended to
evolve a procedure of successive refinement of trackline
spacing. If the initial _rms accuracy were acceptable, the
initial trackline spacing would be accepted and the survey
pronounced completed for that area. If not, the survey
pattern would be tightened by filling in tracklines
half-way between the original ones, with a reserve sample
obtained at a few randomly interspersed quarter-interval
lines. The procedure would be thus iterated until an
acceptable accuracy level was obtained.

The design of an experiment to test the feasibility
of achieving an accuracy of 0.3m in MSS-interpolated depths
must be quite different than that common to the studies

65

conducted and reviewed in the present project. All results
presented here pertain to depth estimates calculated from
Equation (6) using a single set of regression coefficients
for a given site. In the interpolation scheme, the
coefficients must be adjusted continuously according to
local correlations between upwelled radiance and water
depth, as constrained by the corresponding in situ
measurements of K4 and R4; the amount of along-track data
to be included in determining local coefficients must be
evaluated experimentally in the context of a standard
optimal interpolation method. Development of these concepts
in more detail is left for future projects.

The evenness of the bottom and the estimated
accuracy level of the predicted depths would be used to
determine the feasibility of using this technique in each
site on a local basis.

D. COST BENEFIT ANALYSIS

Once an error analysis has justified using MSS or TM
derived depths for interpolating between tracklines,
several factors combine to reduce the cost of the survey.

The principal savings results from a reduction in the
number of days required for nearshore (i.e. within the 100m
depth contour) survey coverage. (This is also the most
dangerous area to survey) . Nearshore surveying (when done
by conventional means) is also very time consuming, despite

66

the relatively small area involved. The greater density of
data required to adequately develop areas within the 100m
depth curve, (500m spacing versus 1000m spacing outside
this zone) requires extensive boat time and processing
time. It is not unusual for the nearshore surveying and
processing to take longer than the offshore survey of much
larger areas. This reduces the "mother" ship to a
support vessel, rather than a data acquisition vessel.

The benefits to be realized from the use of MSS depth
interpolation fields are cost savings which scale directly
as the ratio of reduced survey track miles to survey miles
at standard trackline spacing, less a relatively small
overhead associated with acquisition and analysis of
Landsat MSS imagery and ocean optical properties. A
reasonable estimate for acquiring and analyzing a single
MSS image is $3000.00, and up to five images might be
required for a single 100km X 100km survey site
(considering cloud cover and tidal stage variations) .
Assuming a complete towed optical system per unit cost of
$50,000 and an average useful life of 1000 operating days,
the optical package would add only $50 per day to survey
vessel costs; this is an insignificant increment compared
to (conservatively) estimated costs of order $10,000 per
operating day for Class-I survey vessel. Using these
figures, if a 50 percent reduction in trackline spacing
were achieved through use of MSS depth interpolation in a

67

site requiring ten ship days at conventional spacing, the
cost savings would be approximately $35,000. Assuming only
ten candidate survey sites per year have ocean optical
properties and bottom reflectance characteristics which are
suitable for this technique, the annual savings would
amount to approximately $350,000. At a very plausible
assumed average reduction of 75 percent, the pre-survey and
annual savings would be $60,000 and $600,000. These
estimates are admittedly speculative, but they are
reasonable speculations and show that significant benefits
can be anticipated if MSS depth interpolation accuracies of
approximately 0.3m can be realized.

E. RELATED TECHNOLOGICAL DEVELOPMENTS

Recent advances in remote sensing technology are
expected to improve the accuracy levels of satellite
bathymetric data. The most significant advance in
multispectral imaging technology is the Thematic Mapper
(TM) aboard Landsat D. This system offers improved spatial
resolution (30m versus 96m using the MSS) which will reduce
the smoothing of bottom features over each field of view.
Spectral resolution has also been improved from the MSS
0. lMrr\ band to approximately a O.OlAxm. Therefore, the
wavelength dependent terms in the optical model are apt to
be more representative of the radiance sensed by a
particular band. Additionally, the TM carries two bands

68

suitable for depth prediction: one, centered at 0.48#m,
(closely corresponding to the maximum transmi ttance of
relatively clear water) and the other centered at 0.58/{m.
This redundancy can be used to compute pixel by pixel
values for the bottom reflectance, as well as the water
depth, which was previously possible to only a few m with
the MSS Band 5. Also, depths may be computed with greater
digital resolution because of the instrument's finer
radiometric resolution. These improvements, combined with
in situ optical measurements, should ensure predicted
depths accurate and reliable enough to serve as
interpolation data in many circumstances.

The development of. the Multispectral Active/Passive
Scanner (MAPS) can provide an alternate source of water
depth data. This system may be mounted on fixed wing
aircraft and has the advantage of providing its own optical
ground truth. While the passive scanner records reflected
radiance in several blue-green bands, a pulsed laser (also
in the blue-green wavelengths), is used to determine the
optical properties of the water column and bottom. With
these param, the passive data can be used to derive
information to depths as great as 40m (DMAHTC, 1984).

Other applications of laser technology include the
Hydrographic Airborne Laser Sounder (HALS) , another
aircraft system, which can provide high density (9m
intervals) sounding data across a swath 200m wide. This

69

system has demonstrated the capability to measure depths
within IHO accuracy standards (DMAHTC, 1984). This system
might be used to quickly acquire the control depths
necessary to evaluate the accuracy of the MSS or TM,
reducing still further the burden on survey launches.

The use of Synthetic Aperture Radar (SAR) imagery in
conjunction with MSS or TM imagery is expected to provide
more detailed bathymetric information for pre-survey
analysis and planning. SAR is an active sensing system
operating in the microwave range. It has been observed
that certain surface features observed on SAR imagery
(surface gravity wave refraction patterns, internal waves,
and anomalous backscatter patterns) are tied to distinctive
bathymetric features (Lodge, 1983). Because SAR is an
active system, it's operation is not dependent upon weather
conditions. This can be used to great advantage when
compiling pre-survey bathymetric charts in each area where
the use of MSS or TM data is often limited (e.g., in the
sub-tropical convergence zones which are often
cloud-covered) . The potential application of SAR imagery
and multispectral data used jointly is further discussed in
Chapter V (Recommendations).

70

V. CONCLUSIONS AND RECOMMENDATIONS

From the information presented in the preceding
chapters, it is clear that quantitative bathymetric maps
can be prepared from Landsat MSS images of hydrographic
survey sites characterized by transparent waters and a
highly reflective bottom. Used as an independent data
source, with regression coefficients fit to control depths
measured at only a few locations, rms errors typically
ranged from 1 to 2.5m under appropriate optical conditions.
The primary source of error is spatial variability in ocean
optical properties and/or bottom reflectance. This accuracy
level is adequate for use of MSS imagery as a tool in
bathymetric chart review and revision (Section IV.C.l), and
for pre-survey planning (Section IV. C. 2), but falls far
short of the 0.3m accuracy standard for soundings less than
20m (Umbach, 1976). The conclusion is inescapable that
bathymetric charts derived from remotely sensed
mult ispectral images alone are far too inaccurate, even
under ideal optical conditions, to obviate the need for
conventional in situ hydrographic surveys (Section IV. C. 3).
However, significantly improved accuracy may be expected if
MSS images are instead used to interpolate depths between
concurrent ship survey tracklines at wider-than-normal
intervals. Further experimental research is needed to
determine whether 0.3m depth accuracies can be realized,

71

but if so, the MSS interpolation technique will prove to be
extremely cost-effective. The following sections outlined
recommended procedures for integrating Landsat MSS-derived
depth products into a bathymetric charting and surveying
operational demonstration program.

A. RECOMMENDED CHART REVISION AND SURVEY PLANNING

PROCEDURES

Figure °i illustrates schematically a recommended
sequence of procedures through which MSS images may be
routinely incorporated into long-range hydrographic
resource planning and interim chart revision. The
construction of the preliminary bathymetric "working chart"
would first require obtaining a combination of
mul tispectral (MSS now and TM when available) and SAR
imagery for each area in which DMAHTC is the responsible
authority. As indicated, data for each area would be
acquired in order of priority. This imagery would be used
to depict relative depth zones and to delineate dangerously
shallow waters, as clearly as possible. The use of SAR
imagery is recommended for identifying areas which exhibit
manifestations of shoals; these areas can then be more
carefully investigated using the Landsat imagery. The
working bathymetric chart should also contain annotation
pertaining to suspected hazards or anomalies which could

72

(B)

(A) OBTAIN SAR/MSS/TM
DATA FOR AREAS OF
INTEREST (AS INDICATED
BY PRIORITY LIST)

PROCESS AND INTERPRET IMAGES

TO PREPARE "WORKING
BATHYMETRIC CHARTS"

(C) COMPARE "WORKING CHARTS"
WITH EXISTING NAUTICAL CHARTS

(D) CATEGORIZE AREAS
BY QUALITY OF
NAUTICAL CHART
COVERAGE AND PREPARE
LONG-RANGE HYDROGRAPHIC
SURVEY SCHEDULES

(E) PERFORM INTERIM

CHART REVISION

(ASSIGN "UNCONFIRMED

STATUS" TO SUSPECTED

HAZARDS DISCOVERED

ON IMAGERY)

(F) PERFORM PRE-SURVEY
ANALYSIS AND

PLANNING
(SEE FIGURE 10)

Figure 9. Use of Mult i spectral Imagery in Chart
Revision and Long-Term Planning

73

not be clearly interpreted or delineated with either the
multispectral or SAR imagery.

As shown in Step C, the completed working chart is
intended to offer a basis for evaluating how well the
available chart coverage represents the current bottom
topography of an area. This evaluation would provide input
to planners to further categorize areas of equal priorrty.
In Step D, this additional input will ensure that those
high priority areas most lacking in quality hydrographic
data receive first attention. The information appearing on
this working chart would also be used to update the current
nautical charts, until the area had been properly
re-surveyed. The use of "uncomf irmed hazard" and "position
approximate" labels would differentiate this information
from controlled survey data while still warning mariners of
possible dangers.

The bathymetric working chart and updated nautical
charts are also intended to be useful tools in the
pre-planning phase of a survey, as illustrated by the
arrows leading from Steps B and E to Step F. This concept
is more fully developed in the next section.
B. BATHYMETRIC SURVEY ENHANCEMENT USING MSS DEPTH

INTERPOLATION

Figure 10 illustrates a recommended sequence of
procedures for utilizing MSS images as depth interpolating
fields in bathymetric surveys in sites shown by pre-survey

74

(A) OBTAIN BATHYMETRIC

"WORKING CHART" AND
REVISED NAUTICAL CHART

(B) PLAN SURVEY (LAY OUT
OPTIMAL AND NORMALLY
■ SPACED BOAT TRACKLINES;
ALLOCATE RESOURCES TO OBTAIN

MSS/TM GROUNDTRUTH FOR
ANALYSIS, AND TO INVESTIGATE
"UNCONFIRMED" HAZARDS IDENTIFIED
ON "WORKING CHART"; REQUEST
MSS/TM COVERAGE)

i .

(C) OBTAIN CONCURRENT
MSS/TM COVERAGE

r

(D) OBTAIN CONTROL DEPTHS
AND OPTICAL DATA

(E) PERFORM
ERROR ANALYSIS

I

ACCEPTABLE

ERROR LEVEL

UNACCEPTABLE
ERROR LEVEL

(F) PERFORM SURVEY USING

CONCURRENT MSS/TM
COVERAGE, IN-SITU OPTICS,
AND OPTIMAL TRACKLINE
SPACING

I
(G) PROCESS CONVENTIONAL
SURVEY DATA AND INTERPOLATE
BETWEEN TRACKLINES USING
MSS/TM-DERIVED DEPTHS

(H) PERFORM CONVENTIONAL
SURVEY WITH NORMAL
TRACKLINE SPACING

(I) PROCESS CONVENTIONAL
SURVEY DATA

Figure 10. Integration of Multispectral Imagery into a
Hydrcgraphic Survey Program

75

analysis to be suitable in terms of ocean optical
properties and bottom reflectance. The basis for this
approach was discussed in Section IV. C. 4.

For favorably rated areas, the first step, shown in
Step A, is to obtain the preliminary working chart and the
revised nautical chart, which would have already been
prepared. These would be used, in Step B, to plan the
nearshore survey.

Step B (Figure 10) encompasses three major planning
tasks. First, two sets of survey tracklines would be
prepared for the nearshore areas: one set (designed to
take advantage of the depth matrix supplied by MSS or TM
imagery) would be spaced at wider intervals than those used
in conventional surveys; a second normally spaced set,
would also be prepared in the event the error analysis is
Step E produced unacceptable error levels.

The second planning step would involve the allocation of
resources for those tasks other than regular trackline
surveying. This amounts to investigating those unconfirmed
hazards detected in the imagery, or previously reported by
mariners, and obtaining optical measurements and
groundtruth (Step D) for the analysis in Step E. The
information depicted on the working chart should make
locating potentially hazardous areas more efficient and
thorough. Not only is it more likely that dangerous areas
would be properly investigated, but the entire nearshore

76

survey operation would be safer with more of these areas
identified ahead of time.

The third step involved in the planning phase would be
to request MSS or TM imagery to be taken at the time
scheduled for acquisition of ground truth and optical
measurements. Scheduling for this task would be driven by
the 18-day coverage cycle of the Landsat satellite, by the
availability of boats and crew to obtain groundtruth, and
by weather conditions. Steps C and D represent the
simultaneous acquisition of all data required to perform a
thorough error analysis in conjunction with the survey.
Soundings and optical properties might best be measured by
a single boat dedicated to this task. Before the survey
proceeded from one site to the next, this boat could be
sent ahead to acquire the necessary data. The sooner the
data are in hand the more easily necessary changes could be
planned before actual survey operations were well underway.
With modern microcomputers, it is possible to take optical
measurements using towed instruments, with minimal operator
interface; a two or three man crew could easily obtain the
necessary data sample.

The analysis to be performed in Step E would be similar
to that described in Section III.C. First the depth
measurements would be randomly divided into two samples. A
test sample would be used to derive the coefficients A and
B (from Equation 6) required to predict depths. From

77

these, values for K4 and RD4 would be derived and compared
to the in situ optical measurements; if necessary, the
coefficients would be locally adjusted to these
measurements. The resulting coefficients would be used to
predict depths for each X value (computed from the Landsat
counts per Equation 2) in the control sample. An
unacceptable error level in Step E would automatically
require switching to conventional survey procedures. If
the error level were acceptable, nearshore survey
operations would proceed with boats following widened
survey tracklines and carrying towed instruments for
measuring water clarity.
C. RECOMMENDATIONS

In summary, the results of the literature review and
case studies performed in this thesis project lead to four
key recommendations related to use of visible wavelength
mul t ispectral imagery in an operational bathymetric mapping
program:

1. Procedures should be implemented for routinely
using Landsat MSS images in the planning and management of
operational bathymetric survey and charting programs. The
benefits to be derived from interim chart revision and
pre-survey and pre-survey planning using MSS and
SAR-derived depth manuscripts are well worth the small
expenditure of resources required for processing and
compi lat ion .

78

2. Research should be carried out to develop a
quantitative procedure for determining "lowest probable
bottom reflectance" values for use in all MSS depth
algorithms. The motive for this procedure is to bias
errors towards safe underestimates, and to reduce the
probability of dangerous depth overestimates to a low
value. A primary objective of experimental research in
this area should be to quantify the probability of depth
overestimates exceeding 0.3m (the accuracy standard for
depths less than 20m) and establish associated confidence
levels .

3. New experimental research should be initiated to
determine accuracies associated with using an MSS image as
a basis for interpolating depths between in situ survey
tracklines, with spatial trends in ocean optical properties
constrained by in situ measurements along each trackline.
Existing data from previous experiments are not properly
distributed for this purpose, although an initial
assessment could be performed by modifying the present case
study of the Florida Test Area to interpolate between
charted survey depths at varying trackline spacing. As
was discussed in Section IV. D, significant cost savings can
be anticipated even if only ten or so survey sites per year
have ocean optical properties and bottom reflectance
suitable for use of MSS depth interpolation; the potential

79

savings are large enough to justify placing a high priority
on quantitative evaluation of this particular approach.

4. In contrast to the promising potential for depth
interpolation using MSS images, all results of the present
study suggest that MSS derived depths are not sufficiently
accurate to serve as a direct, independent basis for
bathymetric mapping. Due to spatial variability in
optical properties, error levels lm (or more) greater than
the 0.3m allowable level are expected. Therefore, it is
recommended that research and development efforts
de-emphasize (or even abandon) this extremely demanding
goal and focus on the more plausibly achieved applications
emphasized above.

Finally, it is recommended that future work in this
topic area be organized with a formally established
technical development and operational demonstration
program. Obviously, many of the technical questions implied
in the above research can be approached through
uncoordinated, stand-alone experiments. To convincingly
establish the cost effectiveness and operational utility of
these techniques, however, a comprehensive program of
carefully designed demonstration surveys is considered
essential.

80

APPENDIX A. LANDSAT SATELLITE SERIES AND THE MULT ISPECTRAL
SCANNER

The following general discussion of the Landsat
Satellite series and the Multispectral Scanner (MSS)
applies to Landsat-1 through Landsat-3. Landsat-4,
launched on July 16, 1982, is an operational satellite,
whereas the first three were experimental. Furthermore,
Landsat-4 has different orbital characteristics and imaging
systems than the previous satellites, and will require more
complex processing techniques.

1. LANDSAT SERIES - CHARACTERISTICS

The Landsat series of satellites has served primarily as
a research and development tool to demonstrate the
usefulness of remote sensing from space to various earth
science applications (Warne, 1978). Landsat-1 (initially
named ERTS-1) and its two successors, Landsat-2 and
Landsat-3, were launched in 1972, 1975, and 1978,
respectively; the durability of each satellite has resulted
in nearly continuous coverage since the initial launching
in 1972 (Taranik, 1978). Although post-processing
techniques have generally lagged behind the remote imaging
technology, the variety and quality of users has continued
to increase with each mission. Landsat-4 has incorporated
technology designed to accommodate many of the common user

81

requirements which have developed through exploitation of
data from earlier missions (Taranik, 1978).

Each satellite was launched into a near-polar,
sun-sychronous orbit, at an average altitude of 900km. The
angle between the sun, center of the earth, and the
satellite is maintained at 37.5 degrees, and the orbital
plane is inclined 99 degrees, as measured clockwise from
the equator. These orbital characteristics are designed to
insure repetition of suitable sun-illumination conditions.
The average period of orbit is approximately 103 minutes,
offering repetitive ground coverage, by individual
satellites, of each 185 by 185 km area, every 18 days
(Taranik, 1978) .

2. LANDSAT MULTISPECTRAL SCANNER

Each satellite was equipped with a Multispectral
Scanner (MSS) and a Return-Beam Vidicon (RBV) camera system
as part of its imaging payload. The MSS contains six
detectors in each of the following four wavelength bands:
BAND RANGE (M) TYPE OF RADIATION

4 0.5-0.6 blue-green

5 0.6-0.7 visible red

6 0.7-0.8 invisible reflected IR

7 0.8-1.1 invisible reflected IR

82

(An additional 5th band, which detects thermal infrared
radiance, was installed in the Landsat-3 MSS) . The RBV
camera system has not been widely applied to bathymetric
mapping and so will not be discussed further here. Details
of the RBV camera system are presented in Taranik (1978)
and Slater (1980) .

Each detector in MSS Bands 4-7 samples the reflected
radiance over a 79 by 79m area every 9.95 milliseconds
through a sweep angle of 11.6 degrees; 390 scans (of each
of the six detectors) are formatted into one 34,225 square
kilometer scene, for each of the four bands. (USGS, 1979)

3. SIGNAL PROCESSING TO PRODUCE COMPUTER COMPATIBLE
TAPES

The following discussion refers to pre-1978 procedures,
utilizing the NASA Data Processing Facility (NDPF). With
the introduction of NASA's Image Processing Facility (IPF)
in 1978, other radiometric and geometric corrections were
applied in the production of standard high density computer
Compatible Tapes (CCT's).

The reflected radiance received by each detector is
initially registered as a measure of voltage. The Landsat
MSS can record radiance in either a low gain (normal) or a
high gain setting. In the high gain mode, each signal in
Band 4 and 5 is amplified by a factor of three, which
effectively reduces discrimination between the higher

83

radiance levels while increasing the resolution in the
lower levels (USGS, 1979).

The analog data in each band (except Band 7) is then
"compressed" so that each voltage level is assigned a value
on a scale ranging from 0-63 (see diagram) ; this produces a
quantization which more nearly matches the inherent MSS
photomultiplier noise in each band (USGS, 1979). The
algorithm applied is non-linear, and in effect, results in
a compression of the higher radiance levels and an
expansion of the lower radiance levels.

High gain is preferred for water depth determination. In
water depths greater than 10m, bottom reflectance
contributes only small radiance anomalies above the deep
water radiance L^ 4 (Ch. II). The improved radiometric
resolution at low radiance levels is virtually essential if
these anomalies are to be detected.

Upon reception, the compressed digital values are
decompressed to fit a linear scale ranging from 0-127
counts. Because of the non-linearity of the compression
algorithm, and because of the inherent differences between
compressing an analog signal into a digital value and the
subsequent decompression of that digital value, the lower
level signal values become compacted into relatively few
discrete "steps", or counts, with unused counts at the
higher levels (referred to as forbidden numbers) (Warne,
1978) .

84

Prior to 1978, two additional radiometric corrections
were made, by NASA, to correct for variations in the
response of each detector in each band (Lyzenga and Polcyn,
1979) .

In addition to radiometric corrections, bulk processing
at the NASA Data Processing Facility included geometric
corrections to; 1) normalize the length of the scan lines,
2) register the four bands to one another, and 3) correct
for variations in satellite altitude and attitude as
recorded by on-board equipment (USGS, 1982). Each
corrected image was stored on 800 or 1600bpi
"band-interleaved by pixel pair" format. (This description
applies to pre-1980 Landsat MSS data; procedures and
formats have been modified since then) .

85

APPENDIX B. NAVIGATION OF LANDSAT SCENES

Navigation of imagery produced by a scanning satellite
sensor, such as the Landsat MSS, refers to the development
of an algorithm for assigning map coordinates (e.g.
latitude and longitude) to image pixels. Selection of a
suitable navigation technique is dependent upon 1) the
accuracy required from the algorithm, 2) the availability
of resources, particularly suitable processing facilities
and appropriate firmware and software support, and 3) time
constraints. High accuracy standards require CCT's which
have been geometrically corrected for distortion arising
from rotation beneath the satellite, spacecraft velocity
changes, satellite altitude and attitude variations and
scan skew. The most accurate methods involve modeling the
instantaneous sensor position by measuring the displacement
of image identifiable ground control points from their true
positions on an appropriate projection; from these
measurements, a mapping function is derived which allows
geographic coordinates to be "mapped" onto the input image.
The input image is then re-sampled (within the vicinity of
the mapped position coordinates) to assign a value to the
mapped position of the particular coordinates, since the
position of the coordinates on the input image will not
usually coincide exactly with the center of any one pixel.

86

Although the above procedure is the most accurate,
Landsat tapes which were produced by bulk processing were
not corrected in this fashion. Instead, the bulk processor
utilized on-board monitors to record detectable variations
in the altitude and attitude of the Landsat platform;
geometric corrections are then automatically made in
response to this data. Two further geometric corrections
are made to a) normalize the scan line lengths, and b) to
co-register the four bands. The accuracy of this procedure
is sufficient for the present application.

The navigation procedure used here, models the change in
latitude and longitude from one pixel to another along each
scan line, and from subsatellite point to subsatellite
point between different scan lines. Based upon the known
coordinates of the control points and the subsequent
calculation of the sub-satellite points of the scan lines
in which each control point was located, the change in
latitude and longitude, from any control point to its
scan-line subsatellite point, and thence to any point on
any scan line could be computed. The procedure was based
upon a plane sailing approximation, i.e., that the areas of
concern in each of the two scenes could be treated as flat
surfaces for the purposes of computing differences in
latitude and longitude. This is a generally accepted
assumption for marine navigation problems involving only
small changes in latitude and departure over relatively

87

short distances (under a few hundred miles) (Bowditch,
1977). Computation of the latitude of the point of arrival
can be computed by a direct solution of the plane triangle;
the computation of longitude change requires scaling by the
cosine of the mid-latitude (along which all changes in
longitude are assumed).

The application of this procedure to Landsat MSS images
is described below.

Step 1.

As an initial evaluation, to determine whether or not
the bulk processed tapes of the Bahamas and Florida areas
were sufficiently corrected for use with the proposed
navigation algorithm in the project areas, a plot was made
of the subsatellite points of each scan line containing a
control point. Over the distances covered by each test
area (approximately 15 nautical miles north-south in the
Bahamas test area, and 37 nautical miles north-south in the
Florida test area), it was assumed that the subsatellite
points, for those portions of each scene actually used,
would lie along a straight line (as plotted on a Mercator
projection) .

The maximum deviation from a "best-fit" line for any of
the plotted subsatellite points, in both Bimini and
Florida, was well within the error level associated with
measuring control point position and the subsequent
plotting of subsatellite points.

88

The following equations were used to calculate the
latitude and longitude of the subsatellite point (P0 ) of
each scan line ( X) containing a control point:

LAT(POjAo)-LAT(Pc,*0 tdL(Pc-Po) SlN05/vT } (10)

L0MS(PO)fto) =L0H6(?c,Hc)-(d ( Pc- Po)/COSL3C0j 0s*r , (11)

where: 1) all latitude and longitude values are in minutes
of arc, 2) d(pc-po) is the distance in nautical miles
between the pixel containing the control point and the
subsatellite point, 3) L is the mid-latitude of the portion
of the Landsat scene in which control points lay, and 4)
0SAT i-s tne inclination of the scan line of the MSS aboard
the Landsat-1 satellite, as measured counter-clockwise from
the equator.

The control points (and hence the subsatellite points)
were selected to extend as far as possible across that
portion of the scene from which depth information would be
taken. It was not necessary to select points from over the
entire scene, since any geometric inaccuracies within the
scene would be expected to increase with distance. It was,
however, necessary to extend the selection of the control
points beyond the boundaries of the testing areas (less
than '1 nautical mile in either case) to procure a
sufficient number of reliable points.

89

Step 2.

To test the accuracy of the navigation algorithm, the
"ground control set" from each scene was further divided
into two sample sets: a) a test sample set containing
those points from which the coordinates of the reserve
sample points would be calculated and b) a reserve control
sample with which to test location accuracies. Given the
line number and pixel number of each reserve sample point,
coordinates were calculated from each of the two ground
control points within the sample as:

LAT(P,JO"LAT(po,J0- dK?-Po)SiN05ATl (12)

LONG(P,J0 = L0N6(P0)A>[d(P- Po)/C05 L]C05 0^T ^ (13)

where (P0,£>j is the subsatellite point of the scan line
containing the reserve point (calculated in an intermediate
step) .

In both equations, the first term is the previously
calculated latitude/longitude of the control sample point,
the second term accounts for the change in subsatellite
latitude/longitude with change in scan line, and the last
term then accounts for the change in latitude/longitude
with displacement along a particular scan line.

90

The circular errors in location (at the 90 percent
level) (Bowditch, 1977), were 113.3m based on six reserved
control points for the Bahama Test Area, and 143.5m based
on five reserved control points for the Florida Test Area.

91

LIST OF REFERENCES

Austin, R. W. 1974. The remote sensing of spectral
radiance from below the ocean surface. Opt ical
Aspects of Oceanography. (Jerlov, N. G., and
E. S. Nielson, editors). London: Acedemic Press.

Bowditch, N. 1977. American Practical Navigator (Vol I).
Suitland: Defense Mapping Agency Hydrographic Center,
1386 p.

Brown, W. L., F. C. Polycn, A. N. Sellman, 1971. Water

Depth Measurement by Wave Refraction and Mul t ispectral
Techniques (Report No. 31650-31T). Ann Arbor: Willow
Run Laboratories, 54 p.

Cox, C. and W. Munk . 1954. Measurement of the roughness
of the sea surface from photographs of the sun's
glitter. Journal of the Optical Society of America,
4_4:838 - 850.

Defense Mapping Agency Hydrographic/Topographic Center.
1980. Organization and Function Manual, Washington,
DC.

Defense Mapping Agency Hydrographic/Topographic Center.
1984. Panel presentation delivered by Dr. Kenneth
Daugherty, Technical Director DMAHTC, at Pacific
Conference on Marine Technology, Honolulu.

Gordon, H. R., D. K. Clark, J. W. Brown, 0. B. Brown,

R. H. Evans, and W. W. Broenkow. 1983. Phytoplank ton
pigment concentrations in the Middle Atlantic Bight:
Comparison of ship determinations and CZCS estimates.
Applied Optics. 22:20.

Halperin, M., et al. 1955. Tables of percentage points
for the studentized maximum absolute deviation in
normal samples. American Statistical Association
Journal. 50(269):185 - 195.

Hammack, J. 1977. Landsat goes to sea. Photogrammetr ic
Engineering and Remote Sensing, 43(6):683 - 691.

Hammack, J. 1981. Personal Communication.

Hasell, P. G. Jr., et al. 1977. Active and Passive Multi-
spectral Scanner for Earth Resources Applications on
Advanced Applications Flight Experiment, Report
No. 115000-49-F . Ann Arbor: Environmental Research
Institute of Michigan.

92

Jerlov, N. G. 1976. Marine Optics. Amsterdam: Elsevier
Scientific Publishing Company, 231 p.

Joy, R. T. 1981. Proposed Standard Operating Procedures
for Preparing Landsat Zoned-depth Manuscripts in
Support of NAVOCEANO Pre-survey Planning. Unpublished
report, Washington, DC: Defense Mapping Agency,
Hydrographic/Topographic Center, 21 p.

Lodge, D. W. S. 1983. Expressions of bathymetry on SEASAT
Synthetic radar images. In: Satellite Microwave •
Remote Sensing (T. D. Allan, ed ) . Chichester:
Ellis Horwood, LTD, 526 p.

Lyzenga, D. R., and F. C. Polycn 1979. Techniques for the
Extraction of Water Depth Information from Landsat
Digital Data. Ann Arbor: Environmental Research
Institute of Michigan, 57 p.

Lyzenga, D. R. 1981. Remote Sensing of Bottom Reflectance
and Water Attenuation Parameters in Shallow Water Using
Aircraft and Landsat Data. International Journal
of Remote Sensing. 2(1):71 - 82.

NASA/Goddard Space Flight Center. 1982. Personal
Communications with Landsat project personnel.

Polcyn, F. C, and D. Lyzenga. 1975. Remote Bathymetry and
Shoal Detection with ERTS (final report) . 54 p.
Ann Arbor: Environmental Research Institute of
Michigan.

Polcyn, F. C. 1976. NASA/Cousteau Ocean Bathymetry

Experiment (final report) . Ann Arbor Environmental
Research Institute of Michigan, 132 p.

Polcyn, F. C, D. R. Lyzenga and, F. J. Tanis 1977.

Demonstration of Satellite Bathymetric Mapping (final
report) . Ann Arbor: Environmental Research Institute
of Michigan.

Slater, P. N. 1980. Remote Sensing. Reading : Addison-Wesley
Publishing Company. 547 p.

Taranik, J. V. 1978. Characteristics of the Landsat
Mult ispectral Data System. Sioux Falls: U.S.
Geological Survey, 76 p.

U.S. Geological Survey. 1979. Landsat Data User's
Handbook . Sioux Falls: U.S. Geological Survey.

93

Umbach, M. J. 1976. Hydrographic Manual. Rockville:

National Oceanic and Atmospheric Administration, 307 p.

Warne, D. K. 1978. Landsat Image Analysis: Application to
Hydrographic Mapping. Ph. D. Thesis, Australian
National University, Canberra, ACT., Australia, 217 p.

94

INITIAL DISTRIBUTION LIST

No .Copies

1. Defense Technical Information Center 2
Cameron Station

Alexandria, VA 22314

2. Library, Code 0142 2
Naval Postgraduate School

Monterey, CA 93943

3. Chairman (Code 68Mr) 1
Department of Oceanography

Naval Postgraduate School
Monterey, CA 93943

4. Chairman (Code 63Rd) 1
Department of Meteorology

Naval Postgraduate School
Monterey, CA 93943

5. Dr. James L. Mueller (Code 68My) 3
Department of Oceanography

Naval Postgraduate School
Monterey, CA 93943

6. Richard T. Joy 3
DMA Hydrographic/Topographic Center

Code TOCT
Washington, DC 20315

7. Director Naval Oceanography Division 1
Naval Observatory

34th and Massachusetts Avenue, NW
Washington, DC 20390

8. Commander 1
Naval Oceanography Command

NSTL Station

Bay St. Louis, MS 39522

9. Commanding Officer 1
Naval Oceanographic Office

NSTL Station

Bay St. Louis, MS 39522

10. Commanding Officer 1
Fleet Numerical Oceanography Center

Monterey, CA 93940

95

11. Commanding Officer

Naval Ocean Research and Development

Activity

NSTL Station

Bay St. Louis, MS 39522

12. Commanding Officer

Naval Environmental Prediction
Research Facility
Monterey, CA 93940

13. Chairman, Oceanography Department
U.S. Naval Academy

Annapolis, MD 21402

14. Chief of Naval Research
800 N. Quincy Street
Arlington, VA 22217

15. Office of Naval Research (Code 420)
Naval OCean Research and Development
Activity

800 N. Quincy Street
Arlington, VA 22217

16. Scientific Liaison Office
Office of Naval Research

Scripps Institution of Oceanography
La Jolla, CA 92037

17. Library

Scripps Institution of Oceanography

P. 0. Box 2367

La Jolla, CA 92037

18. Library

Department of Oceanography
University of Washington
Seattle, WA 98105

19. Library
CICESE

P. 0. Box 4803

San Ysidro, CA 92073

20. Library

School of Oceanography
Oregon State University
Corvallis, OR 97331

96

21. Commander

Oceanographic Systems Pacific

Box 1390

Pearl Harbor, HI 96860

22. Director

Defense Mapping Agency

Code PPH

Bldg. 56, U.S. Naval Observatory

Washington, DC 20305

23. Director

DMA Hydrographic/Topographic Center

Code HO

Washington, DC 20315

97

13 37 5

210277

Thesis

J81+2

Joy

cl

An assessment of the

potential role of multi-

spectral imagery in

bathymetric charting.

210277

Thesis

J8U2

cl

Joy

An assessment of the
potential role of multi-
spectral imagery in
hathymetric charting.