Geophysical Prospection at Portus:
An Evaluation of an Integrated Approach to
Interpreting Subsurface Archaeological Features
Prepared By: Jessica Ogden
A Dissertation submitted in partial fulfillment of the degree of:
MSc Archaeological Computing in Spatial Technologies
Instructional Course
University of Southampton
School of Humanities
Department of Archaeology
2008
1
Table of ContentsList of Tables
List of Figures
List of Maps
Abstract
Acknowledgements
......................................................................Chapter One: Introduction! 12
..................................................................................................1.1 The Archaeology! 12
.......................................................................................................1.2 The Research! 13
................................................................................Chapter Two: Portus ! 14
....................................................................................2.1 Portus: A Brief Summary! 14
..........................................................................................................The Claudian Harbor! 14
...........................................................................................................The Trajanic Harbor! 15
..................................................................2.2 Previous Work: Towards Integration! 16
......................................................Chapter Three: State of Geophysics ! 19
................................................................................3.1 Archaeological Geophysics! 19
............................................................................3.1.1 The Possibilities and Motivations! 19
..........................................................................................................3.1.2 The Limitations! 19
.................................................................................................................Uncertainty! 19
..............................................................................................Issues of Interpretation! 20
..................................................................................................Physical Constraints! 20
......................................3.2 Geophysical Data Integration: A Difference in Terms! 21
.....................................................................................3.2.1 Integrated Survey Methods ! 21
.......................................3.2.2 Using Geographic Information Systems for Integration! 22
..........................................................................................3.2.3 Integrated Data Analysis! 22
....................................................................Chapter Four: Case Studies ! 24
..........................................................4.1 Multiple Approaches to Data Integration! 24
...................................................................................4.1.1 “Digital Image Combination”! 24
........................................................................................4.1.2 “Quantitative Integration”! 25
.................................................4.1.3 A Synthesis of Integration Techniques: Army City! 26
........................................................Chapter Five: Research Objectives! 28
1
.......................................................................Chapter Six: Methodology! 30
........................................................................6.1 Data Collection and Processing! 30
...........................................................................................................6.1.1 Magnetometry! 30
..................................................................................................................The Survey! 31
...........................................................................................6.1.2 Resistance Tomography! 31
.....................................................................................................................6.1.3 Augering! 33
...................................................................................................................6.1.4 Resistivity! 33
.......................................................................................6.1.5 Ground Penetrating Radar! 34
.................................................................................................6.2 Data Preparation! 43
..................................................................................................6.2.1 Migration to the GIS! 44
...........................................................................................6.2.2 Preparation for Analysis! 45
....................................................................................................Achieving Normalcy! 45
..............................................................................................Binary Data Generation ! 46
.......................................................................................................6.3 Data Analysis! 46
..........................................................................6.3.1 Traditional Interpretation Methods! 46
........................................................Digitization of Interpretations and Initial Results ! 46
.......................................................................................6.3.2 Graphical Data Integration! 47
.........................................................................................Two Dimensional Overlays! 47
..................................................................................................Translucent Overlays ! 47
................................................................................................RGB Color Composite! 48
............................................................................................................3D Integration ! 48
.............................................................................................6.3.3 Discrete Data Analysis! 50
..................................................................................................Binary Data Analysis ! 50
..........................................................................................................Cluster Analysis! 51
........................................................................................6.3.4 Continuous Data Analysis! 54
...............................................................................................Data Sum and Product! 54
......................................................................................................Data Max and Min! 54
.................................................................................Principle Components Analysis ! 55
...........................................................................Chapter Seven: Results! 56
........................................................................7.1 Interpretation of Survey Results! 56
...........................................................................................................7.1.1 Magnetometry! 57
...........................................................................................7.1.2 Resistance Tomography! 58
...................................................................................................................7.1.3 Resistivity! 59
2
.............................................................................................................................7.1.4 GPR! 60
..................................................................7.2 Results of Integrated Data Analysis! 63
....................................................................................................7.2.1 Graphical Overlays! 63
.........................................................................................Two Dimensional Overlays! 63
..................................................................................................Translucent Overlays ! 63
................................................................................................RGB Color Composite! 64
............................................................................................................3D Integration ! 64
.............................................................................................7.2.2 Discrete Data Analysis! 65
...........................................................................................................Binary Analysis ! 65
..........................................................................................................Cluster Analysis! 66
........................................................................................7.2.3 Continuous Data Analysis! 66
...........................................................................................................Data Functions ! 67
.................................................................................Principal Components Analysis ! 67
.......................................................................Chapter Eight: Discussion! 68
.........................................8.1 Some Potential Implications of the Interpretations! 68
....................................................8.2 Complications and Critique of Methodology! 69
.............................................................................................................................8.2.1 GPR! 69
.........................................................................................................Velocity Analysis! 69
............................................................................Correction for Tilt and Topography! 70
...............................................................................................................Edge Effects ! 71
.............................................................................................................8.2.2 Classification! 71
..........................................................................................Binary Data Classification! 71
................................................................................To Be, Or Not To Be Supervised! 72
..............................................8.3 An Assessment of Limitations and Applicability! 72
...................................................................................................8.3.1 Discrete Data Input! 72
......................................................................................8.3.2 The Number of Data Inputs! 73
....................................................................................................8.3.3 The Level of Detail! 73
...................................Chapter Nine: Future Prospects & Conclusions! 75
..............................................................................................Bibliography! 77
....................................................................................Appendix A: Maps! 80
....................................Appendix B: Resistance Tomography Figures! 117
...........................................................................Appendix C: Text Files ! 121
3
............................................................................................................Dendrogram: ! 121
.....................................................................................................PCA Parameters: ! 123
4
List of Tables.................................................................................Table 1: Header File Structure" 34
.............................................................................Table 2: Binary Data Thresholds" 35
.....................................................Table 3: RGB Composite Color to Feature Key" 36
List of FiguresIn the Text:
.................................................................Figure 1: Location Map of Portus in Italy" 1
..................................................................Figure 2: Illustration of Claudian Harbor" 4
...................................................................Figure 3: Illustration of Trajanic Harbor" 5
................................................................Figure 4: Geographic Area of Research" 19
.....................................................Figure 5: Depiction of GPR Traverse Intervalsr" 25
..................................Figure 6: Screen Shot of Range Gain for GPR Processing" 26
......................................................Figure 7: Progression Radargram Processing" 29
............................................................................Figure 8: X-Plane GPR Timeslice" 30
.......................................................................Figure 9: Topo-Corrected Timeslice" 31
...........................Figure 10: Topo-Corrected Timeslice With IsoSurface Render" 32
...........................................Figure 11: 3D View of GPR and Excavation Features" 38
.......Figure 12: 3D View of GPR and Excavation Features With Building Survey" 39
..................................................Figure 13: 3D Illustration of Boolean Operations" 40
...........................................................Figure 14: Results of Query of 3D Features" 54
1
.................................................Figure 15: Proposed Schema of Pylon Locations" 58
..............................Figure 16: Illustration of Velocity Changes Over Topography" 60
In Appendix B:
..............................Figure 17: Resistance Tomography Profiles 1960 and 1988" 106
..............................Figure 18: Resistance Tomography Profiles 2000 and 2008" 107
..............................Figure 19: Resistance Tomography Profiles 2020 and 2030" 108
..............................................Figure 20: Resistance Tomography Profiles 2038" 109
List of Maps............................................................................................Map 1: Magnetometry " 69
...................................................................Map 2: Magnetometry Interpretations" 70
........................................................Map 3: Resistance Tomography Profile Map" 71
.......................................Map 4: Resistance Tomography Interpretation Overlay" 72
.....................................................................................................Map 5: Resistivity" 73
..........................................................................Map 6: Resistivity Interpretations" 74
....................................Map 7: Ground-Penetrating Radar Timeslice A: 0-22 cm" 75
..........Map 8: Ground-Penetrating Radar Timeslice A Interpretations: 0-22 cm " 76
..................................Map 9: Ground-Penetrating Radar Timeslice B: 13-34 cm " 77
......Map 10: Ground-Penetrating Radar Timeslice B Interpretations: 13-34 cm" 78
................................Map 11: Ground-Penetrating Radar Timeslice E: 50-72 cm " 79
......Map 12: Ground-Penetrating Radar Timeslice E Interpretations: 50-72 cm " 80
2
................................Map 13: Ground-Penetrating Radar Timeslice G: 75-97 cm" 81
.....Map 14: Ground-Penetrating Radar Timeslice G Interpretations: 75-97 cm " 82
..............................Map 15: Ground-Penetrating Radar Timeslice H: 88-110 cm" 83
...Map 16: Ground-Penetrating Radar Timeslice H Interpretations: 88-110 cm " 84
............................Map 17: Ground-Penetrating Radar Timeslice K: 126-147 cm" 85
.Map 18: Ground-Penetrating Radar Timeslice K Interpretations: 126-147 cm " 86
...........................Map 19: Ground-Penetrating Radar Timeslice M: 151-172 cm" 87
.Map 20: Ground-Penetrating Radar Timeslice M Interpretations: 151-172 cm "88
...............................Map 21: 3D View of GPR and Excavation Features: Oblique" 89
.............................Map 22: 3D View of GPR and Excavation Features:Overhead" 90
..............Map 23: Graphical Overlay of Resistivity Contours on Magnetometry" 91
................................................................................Map 24: Translucent Overlay 1" 92
................................................................................Map 25: Translucent Overlay 2" 93
.........................................................................................Map 26: RGB Composite" 94
...........................................................................Map 27: Boolean OR Calculation" 95
.........................................................................Map 28: Boolean AND Calculation" 96
................................................................................................Map 29: Binary Sum " 97
............................................................................Map 30: Binary Sum >2 Methods" 98
........................................................Map 31: Maximum Likelihood Classification" 99
.....................................................................................Map 32: Class Probability" 100
.................................................................................................Map 33: Data Sum " 101
............................................................................................Map 34: Data Product" 102
3
........................................................................................Map 35: Data Maximum " 103
.........................................................................................Map 36: Data Minimum " 104
...........................................................................................................Map 37: PCA" 105
4
AbstractIncreasingly, archaeo-geophysicists have began to take “integrated” approaches through the use of
multiple survey methods to investigate potential archaeological features. However, often when a multi-
method approach is taken, the interpretations, analysis, and presentation of the resulting data is limited
to “side by side” comparisons of gray-scaled graphical representations of the data. A distinction is
made here between integrated survey methodologies and integrated data analysis. Recent
developments in geophysical data analysis have suggested that in addition to a multi-method approach,
“data fusion” techniques can offer meaningful insights into archaeological features, as well as allow for
researchers to establish patterns between multivariate data sets that might otherwise go unnoticed.
This research attempts to compare and contrast a variety of mechanisms for data fusion, and assess
their applicability to the ancient port of Imperial Rome, Portus.
5
AcknowledgementsFirst and foremost, I would like to thank my family. Though this venture has taken me across the globe,
you all have loved and supported me throughout, and I am quite sure none of this would have been
possible otherwise. A very special thank you to Grandpa, for everything you have done for me, this
work is dedicated to you.
This piece of work is the result of a tremendous amount of work put forth by many people. First, thanks
to all of those involved with the Portus Project, it’s been so much fun working with all of you, and I can’t
wait to return next year! Many thanks the Camerone (past and present) that made this possible, thanks
for all the advice, the pushing, pulling, and general good time at Portus. Thanks to Dean Goodman for
all your last minute technical assistance. Thank you to Jean-Philippe Goiran and Ferreol Salomon for
graciously providing me with the results of the coring. Thank you, Simon Keay for championing
geophysics, your enthusiasm for my work was most encouraging. Thanks to David Wheatley for all of
your instruction, and patience. Special thanks to Kris Strutt, I have appreciated your unrelenting advice,
support, and guidance throughout. Thanks to Gareth Beale for the fun times in the field, the pep-talks in
the lab, and the many insightful glimpses into the world of Roman architecture. Many thanks to Graeme
Earl for all the advice and reassurance, but especially for the opportunity to be a part of the Portus
Project. I am quite sure things would be very different if you hadn’t suggested that Kvamme paper.... To
my partner in crime, Sarah: What would I have done without you?
And to Leif, thank you, thank you, thank you. You saw the best and the worst of this research, and
without your patience, understanding, and helpful critique, the outcome would have been very different.
And to all of those who helped me to stay present.
6
Chapter One: Introduction1.1 The Archaeology
The site of Portus, located north of the mouth of the River Tiber, served as the main port of Rome during
the majority of the Imperial Period of the Roman Empire. The initial construction of a harbor at Portus is
believed to have occurred around 42 AD under the reign of Emperor Claudius. (Keay et al. 2005:11)
This construction involved linking the harbor basin to the River Tiber through a series of canals and
aqueducts. Later under the reign of Emperor Trajan, Portus was expanded, potentially to withstand the
increased economic traffic occurring between Rome and the rest of the Empire.
One of the structures erected around the Trajanic Harbor was an extensive complex now known as the
Palazzo Imperiale or “Imperial Palace.” This structure and the surrounding area, situated between the
Trajanic and Claudian Harbors, was the original focus of recent and future excavations as part of the
Portus Project, and is the location of this geophysical survey and research.
Figure 1: Location Map of the Site of Portus (Keay et al 2005)
1
1.2 The Research
Extensive and intensive geophysical prospection has been employed at Portus within recent years,
proving an integral role in discerning the nature and extent of the archaeological record of the port
complex. Recent excavations have allowed for a reciprocal relationship to exist between geophysical
and archaeological research, and have paved the way for a regime of meaningful, integrated geophysical
research at Portus. Many types of geophysical and archaeological survey methods have been
employed, including magnetometry, electrical resistance, ground-penetrating radar, standing building
and micro-topographic survey, and others, to interpret the archaeological record, as well as provide an
immense volume of data to be compared and contrasted to the excavation data. The sheer quantity of
data, as well as the nature of the archaeology have provided an ideal site for the exploration of spatial
data and remote sensing analysis techniques, as well as the assessment of their utility within archaeo-
geophysical research as whole. This research attempts to critically assess the field and data processing
methodologies used, as well as examine the applicability of a variety of mathematical and multivariate
analysis approaches, such as cluster and principal components analysis, to the prospection results at
Portus.
The archaeological context of Portus is presented in Chapter 2, along with previous research strategies
for assessing of the port complex. Next, I have made attempts at describing the geophysical back drop
for which this research occurs, giving an outline of the benefits and limitations of the use of geophysical
prospection within archaeological research in Chapter 3. In Chapter 4 a few supporting case studies are
presented as additional qualifications for my research objectives which are outlined in Chapter 5. The
field and data analysis methodologies are presented in Chapter 6, and all results and discussion has
been left for Chapters 7 and 8.
2
Chapter Two: Portus2.1 Portus: A Brief Summary
The Claudian Harbor
Prior to the construction of the Claudian Harbor at Portus, Rome relied on the shipment of goods
from Puteoli, in the Bay of Naples where goods would either be shipped to Ostia, where an extensive
river port existed, or carried overland to the city. (Keay et al. 2005:297) It is believed that ships could
then anchor and transfer goods to barges at the mouth of the River Tiber at Ostia, and be floated up
stream to Ostia or Rome. (Rickman 1980:18, 46; as cited by Keay et al. 2005:297) Work on the
artificial harbor began in AD 42 by Claudius, and construction took over 20 years to complete. (Keay
et al. 2005:298) The chosen location of the harbor was presumably to take advantage of the sand
dunes that lay between the salt marshes to the east and the sea to the west. (Keay et al.
2005:298-299)
Incomplete archaeological evidence prevents researchers from having a complete understanding of
the layout of the Claudian Harbor, though it is strongly suggested that the establishment of the new
harbor commenced with the construction of two canals of different functions. (De Gaetano and Strutt
2007:6) The function of the Northern Canal is believed to have been to alleviate the flooding of the
Tiber, while the other, the Fossa Traiana provided a route for the shipment of goods to Rome. (Keay
et al. 2005:298) Next the outer, artificial basin was constructed along with the later Darsena and
surrounding buildings in the Neronian period. (Keay et al. 2005:300) The Darsena, or inner harbor, is
believed to have acted as the key area for the control, storage, and regulation of goods for the
Claudian Harbor. (Keay et al. 2005:299-300)
3
Figure 2: Claudian Harbor Schema (Keay et al 2005)
The Trajanic Harbor
The construction and expansion of the Claudian Harbor under Trajan further affirms the increased
traffic to the harbor, and throughout the Roman Empire at this time. (Keay et al. 2005:305) The harbor
complex was substantially increased in size and in storage capacity to withstand the transport of
grain from Alexandria to Rome. (De Gaetano and Strutt 2007:7) The most unique feature of the
Trajanic Harbor is the artificial hexagonal basin, which could to fulfill both a practical role by potentially
separating port functions to each side of the hexagon, as well as an ideological one by making a
communicating a clear statement of Imperial grandeur. (Keay et al. 2005:308-309)
Additional canals were constructed to re-enforce and enhance the transport of goods on barges up-
stream to Rome, as well as a series of warehouses or horrea on each side of the port complex. (Keay
et al. 2005:309-310) These warehouses allowed for additional storage facilities, increasing the
storage capacity to over 90,000 m!, over three times the storage capacity of the neighboring port of
Ostia. (Keay et al. 2005:310)
One of the most perplexing structures of the port is the so-called Palazzo Imperiale. The function of
this immense structure which still stands at 2 stories, is still under debate,1 though it is thought to
have been constructed during the Hadrianic period, as building techniques and brick stamps suggest
4
1 http://www.portusproject.org/romanbackground/trajanic2.html (Accessed 16/11/08)
this dating.2 The land is known to have been in use prior to the construction of the Trajanic Harbor,
however, there is no defining evidence that suggests the existence this grand structure at that time.
One distinct feature of the Late Antique period worth noting was the construction of the so-called
Mura Constantiniane, a defensive wall constructed to protect the port complex. Sometime after the
4th and early 5th centuries AD, substantial structures were enclosed to provide external defenses for
the port. (Keay et al. 2005:291)
Figure 3: Trajanic Harbor Scheme (Keay et al 2005)
2.2 Previous Work: Towards Integration
A variety of approaches to archaeological survey have been taken at Portus and the surrounding area
that reflect the research goals, as well as the nature and scale of the archaeological deposits on site.
Emphasis has been placed on an integration of methods from the onset, with particular attention on
multi-scalar methods for surveying the archaeological record. As part of the Roman Towns Project in
the Tiber Valley, between the years of 1997 and 2004, extensive magnetometry surveys have been
conducted throughout the region of the port complex by the British School at Rome (BSR) in
collaboration with the Universities of Southampton and Cambridge and the Soprintendenza per i Beni
Archeologici di Ostia. (Keay et al. 2005:63)
5
2 Personal communication: Gregory Tucker, cited (Bloch 1947: 100-102; Blake 1973: 289; Lugli 1957: 607) 16/11/
08
As part of the extensive survey, magnetometry was chosen for its quick and efficient data capturing
capabilities, as well as the magnetic susceptibility of buried brick structures which are highly contrasting
to the existing soil types at Portus. (Keay et al. 2005:64) The magnetometry, combined with extensive
field walking, topographic survey and aerial photo interpretation have proven successful in locating and
mapping large landscape archaeological and geomorphological features at Portus and the surrounding
region. (Keay et al. 2005:61-69) The magnetometry surveys revealed considerable new evidence about
the buildings and canals constructed around the Trajanic and Claudian harbors (Keay et al. 2005). The
following are some results from the key areas on site:3
• A more detailed plan of the Palazzo Imperiale and warehouses
• Sections of a defensive wall surrounding the port ( the Mura Constantiniane)
• A major aqueduct whose location was previously unknown
• Detailed plans of buildings to the southwest of the harbor, surrounding the Basilica, as well as along
the canal
• More evidence for land reclamation and property divisions at the junction of the Tiber and the
Fossa Traiana
The Portus Project is the current project funded by the Arts and Humanities Research Council (AHRC) in
collaboration with the Soprintendenza per i Beni Archeologici di Ostia e Porto, and the Universities of
Southampton and Cambridge, and is a flagship project of the British School at Rome (BSR). (De
Gaetano and Strutt 2007) In 2007, as the first phase of the Portus Project, intensive geophysical
prospection began in the area between the Trajanic and Claudian harbors near the Palazzo Imperiale
with the aim of assessing the depth of overburden and features prior to the commencement of
excavation.4 (De Gaetano and Strutt 2007:4) This area of interest was targeted for many reasons, one of
which lies in it’s unique position between the Trajanic and Claudian harbors. (De Gaetano and Strutt
2007:7) Within this area (Figure 4) the magnetometry revealed a complex series of east-west linear
features, and a large sub-circular feature on Side VI of the hexagon. (Keay et al. 2005:103) However, it is
worth noting that in this topographically and archaeologically complex area, the magnetometry revealed
a substantial amount of near surface rubble, complicating and detracting from the interpretations. (Keay
et al. 2005:99) Thus, a targeted resistance tomography survey in conjunction with 28 shallow hand
6
3 Taken from the British School at Rome website on Portus:
http://www.bsr.ac.uk/BSR/sub_arch/extra/BSR_Tiber_Roman_06.htm
4 Work was undertaken by the Archaeological Prospection Services of Southampton (APSS) and The British School
at Rome (BSR)
auger samples was conducted in May-June of 2007 to complement the magnetometry survey. This
survey revealed the location of atleast one harbor mole, as well as extensive structural remains to the
west of the Mura Constantiniane. (De Gaetano and Strutt 2007)
As further discussed in Chapter 7, the magnetometry and resistance tomography surveys have revealed
a great deal about the archaeological remains on Side VI of the Trajanic Harbor, however, have left many
questions, particularly concerning the “Palazzo Imperiale” and the massive “warehouses,” unclear. The
modern trackway bisecting the “Palazzo Imperiale,” as well as the limitations of magnetic survey, placed
constraints on the archaeologists and geophysicists at Portus, and in turn have limited the interpretation
and construction of a chronological sequence for this area of the port. (Keay et al 2008:14) A core
excavation area of 300 m! was opened in 2007 on Side VI, which began the on-going reciprocal
relationship between the geophysical results and the archaeological data needed to verify the overall
interpretations of the port complex as a whole. (Keay et al 2008:9) The excavation at Portus is on-going,
and though the results are currently unpublished, elements of the recovered data have been used to
calibrate and interpret this research.
7
Chapter Three: State of Geophysics3.1 Archaeological Geophysics
Before discussing the aims of this research project, it is necessary to briefly track the use of geophysical
prospection within archaeology, with the aim of assessing the current trends in data integration, as well
as provide the context for the objectives of this research.
3.1.1 The Possibilities and Motivations
“Cultural resource managers have rapidly grasped the power of geophysical methods to quickly, efficiently, and
nondestructively discover and map sites for selective excavation or avoidance, producing greater economy of
time and resources.” (Conyers 2004)
The potential outcomes and motivations for conducting archaeo-geophysical prospection are
important to mention as they determine and inform many aspects of each field methodology, and
need to be clear prior to the onset of any survey, and this one is no exception. The research
questions, archaeology, and time constraints, for example, all inform the survey resolution, survey
extent, and focus of each methodology. For instance, some motivations for conducting a survey
might be:
• To target research areas for possible excavation of known or previously unknown archaeological
resources
• A cost-effective means for defining or avoiding areas with proposed development, based on the
presence or absence of archaeological resources
• An administrative tool for defining and managing cultural heritage resources
• To conduct and further geophysical prospection research in the field of archaeology
3.1.2 The Limitations5
Uncertainty
However, even under the most ideal survey conditions, there is never 100% certainty that
archaeological resources will be successfully located. (Gaffney and Gater 2003:15) Geological
and archaeologically distinct signatures of specific sites require different strategies, and some
8
5 This section stems from a ‘brainstorming’ session in preparation for a paper on the possibilities and limitations of
geophysical prospection in archaeological research in Italy: October 13, 2008. Professor Simon Keay, Steve Kay,
and Kris Strutt are gratefully acknowledged for their input.
degree of forethought. The suspected type of archaeological features, and the research
objectives will always determine the type of prospection method used, and even then, there is no
guarantee that the survey will render the desired results.
Issues of Interpretation
Some limitations of geophysical prospection present difficulties in interpretation, constraining the
types of conclusions that may be made from survey results. Often, archaeological anomalies are
difficult to differentiate between a range of other “causative bodies” (Gaffney and Gater 2003:15),
rendering the distinction between geological and anthropogenic anomalies difficult, and at times,
uncertain. Other interferences, including modern cultural disturbances, rodent burrows, tree
roots, etc. contribute to this “signal-to-noise” ratio. Even so, “one project’s signal may be
another’s noise,” (Kvamme 2006b: 237) re-emphasizing the point that every site is unique, thus
the employed geophysical methodology must suit the research question, as well as the
archaeology and geomorphology of the area.
At times the limits of the employed method and instrumentation prevent the full characterization
of geophysical features. For instance, neither magnetic gradiometry nor resistivity can resolve the
full extent of features below ground surface, thus without additional depth calibrations, the
chronological sequence of archaeological anomalies is often difficult to ascertain.
Ground-penetrating radar results, in themselves, present a range of interpretation challenges,
associated with the complexity and size of the three-dimensional data volume. Making accurate
interpretations of GPR radargrams is arguably the most important aspect of radar prospection,
and often the most difficult to grasp. (Conyers 2004:9) Lawrence Conyers (2006:145) explains
that often, the timing of GPR surveys present numerous interpretation challenges, particularly to
those inexperienced with interpreting GPR data sets. When GPR surveys are conducted prior to
excavations, and the target features are subtle, data interpretation often requires an ‘experienced
eye’ as well as the integration of information from known feature types to ground truth radar data.
Physical Constraints
The constraints of modern survey environments also present limitations on geophysical survey
data sets. Urban areas often prevent the prospection of large areas, leaving the surveyor
constrained to small sample sizes which may not reflect the full extent of the archaeological
context. Depending upon the characteristics of the archaeology, the modern proprietary
boundaries may not coincide with the extent of the archaeological features, sometimes resulting
9
in issues of access, again preventing a complete perspective on subsurface anomalies. In
addition, topographic and other environmental obstacles such as vegetation, may obstruct or
compromise the integrity of the geophysical data results.
3.2 Geophysical Data Integration: A Difference in Terms
“Integration is, after all, an overarching requirement of the multidisciplinary effort that constitutes most archaeological
research.” (David 2001:525)
Almost every limitation encountered within archaeological prospection can be alleviated or avoided by
employing some form of an integrated approach to investigating subsurface features. However, it is
necessary to clarify the different types of integrations, as this term has been assigned to a variety of
comparative data collection and analysis techniques. A distinction is made here between integrated
survey methodologies and integrated data analysis.
3.2.1 Integrated Survey Methods
It is well known in archaeo-geophysical prospection that no “universal detection device” exists to
detect all subsurface features at any given archaeological site. (Gaffney and Gater 2003:55) Different
means of prospection are employed, dependent upon the physical properties of the suspected
subsurface features being observed. Increasingly, archaeo-geophysicists have begun to take an
“integrated” approach by using multiple survey methods to investigate potential archaeological
features. “The location and characterization of sites is best achieved using several detection
methods,” (David 2001:525) as anomalies not indicated by one survey method may be revealed by
another, as well as add complementary knowledge about the subsurface feature. (Kvamme 2003:
439)
Also, depending upon the nature of the archaeological site and landscape, integration may take form
in using a standard geophysical prospection method (i.e. magnetometry, resistivity, conductivity, GPR,
etc) in conjunction with any of the following methods:
• Augering
• Targeted Excavation
• Geochemical Analysis
• Systematic Field Walking
• Aerial Photography Interpretation
• Topographic Survey
10
• Standing Building Survey (if applicable)
In addition to the use of different types of survey methods, integration may also be achieved by
varying the: instrumentation, resolution, alignment, type of array, and survey depth. As with the
choice of survey method, all of these variables must be assessed and applied within the appropriate
archaeological and geological contexts, as well as the bounds of the research questions being asked.
3.2.2 Using Geographic Information Systems for Integration
Geophysical prospection data, like other recorded archaeological data are spatial data sets which
present numerous opportunities for integration and spatial analysis. Some would argue that
Geographical Information Systems (GIS) are the “interface” between archaeological prospection
results and excavation data. (Neubauer 2004:159-166) GIS, in this sense, provides a space for the
integration of other types of survey data (such as those described in Section 3.2.1) to be examined,
compared and contrasted in their real world spatial contexts. Though relational data bases,
statistical, and computer aided design (CAD) packages provide mechanisms for handling simple
spatial data (Neubauer 2004:160), the GIS provides the opportunity to create meaningful (spatial or
otherwise) relationships between data sets, and acts as an arena for solving archaeological problems.
(Chapman 2006, Neubauer 2004:161)
Migrating geophysical results into a GIS is standard “good practice” in archaeological prospection,
and provides an arena for additional integration through spatial analysis techniques. However, if the
interpretations, analysis, and presentation of the resulting data is limited to “side by side”
comparisons of graphical representations of the data, the full potential for analysis of the
multidimensional geophysical data sets will not be met.
3.2.3 Integrated Data Analysis
Recent developments in geophysical data analysis have suggested that in addition to a multi-method
approach, integrated data analysis, sometimes called “data fusion” (Kvamme 2003: 58) must also be
used to extract the maximum amount of archaeological interpretation from geophysical results.
(Kvamme 2003, 2006a, 2006b; Piro et al. 2000; Neubauer et al. 1997, 2002) The researcher must
first “establish the hypothesis that each geophysical method investigates one event, i.e. the presence
of anomalous volumes underground,” then they have the ability to quantify and integrate each set of
geophysical results. (Piro et al. 2000: 204) Integrated geophysical data analysis allows the
geophysicist to establish interrelationships and patterns between multidimensional data sets, and
therefore improve the identification and interpretation of subsurface anomalies. (Kvamme 2006a: 57)
11
The integration of geophysical results allows the geophysicist to “better define position, extension,
depth, thickness, and physical characteristics of any anomalous body within its geological
context.” (Piro et al. 2000: 212)
Nevertheless, an “uneasy relationship” exists between the desire of geophysicists and archaeologists
to produce “visually pleasing” representations of prospection results and the need to push the
boundaries of traditional anomaly interpretation through statistical and spatial analysis. (Gaffney
2008:329)
12
Chapter Four: Case Studies4.1 Multiple Approaches to Data Integration
An integrated approach to data analysis has been applied in several North American pre-historic and
historic archaeological contexts by Kvamme (2003, 2006a, 2006b), and in classical Roman archaeology
in Austria and Italy by Neubauer et al. (1997, 2002), and Piro et al. (2000), respectively. In each case,
the geophysicists applied different approaches (sometimes using multiple analysis techniques) to extract
the maximum level of interpretation and analysis from the archaeo-geophysical record. It is necessary to
outline a three case studies which have set the precedent for this dissertation.
4.1.1 “Digital Image Combination”
In 1997, Wolfgang Neubauer and Alois Eder-Hinterleitner published results of resistivity and
magnetometry surveys conducted on a five hectare portion of the Roman town of Carnuntum, in
eastern Austria, the former residence of Emperor Marcus Aurelius. (Neubauer et al. 1997:179) In
addition to presenting the possible location of the civil town’s forum, they demonstrated two different
mechanisms for using “digital image overlays” as a means for combining and interpreting the results
from multiple prospection methods.
Using the overlapping 80 x 80 meter area surveyed by the magnetometry and resistivity, Neubauer et
al. first combined the images using the four basic arithmetic operations: addition, subtraction,
multiplication, and division. In the results of the ‘addition’ image, the Roman walls, which were
slightly visible in the magnetometry and highlighted in the resistivity, were clearly revealed. (Neubauer
et al. 1997:185) Conversely, the ‘subtraction’ image removed the walls and highlighted the structure’s
floors. (Neubauer et al. 1997:185) This type of image combination allowed for Neubauer et al. to
distinguish between the observed geophysical properties, and assert the presence of four different
types of floor surfaces within the subsurface Roman structure. (Neubauer et al. 1997:185)
In addition to using arithmetic functions, Neubauer et al. used true Red-Green-Blue color composites
to combine prospection results from two different survey methods. By assigning each survey type a
separate channel, the four different floor types identified with the arithmetic operations, again became
clearly visible. (Neubauer et al. 1997:187)
In short, Neubauer et al. demonstrated that subsurface Roman buildings cannot be resolved with
great accuracy by using only one method, magnetometry or resistivity (Neubauer et al. 1997:179),
13
and that using RGB color composites and arithmetic operations to interpret subsurface features from
image overlays produced a meaningful tool for cross-correlating archaeo-geophysical prospection
results. (Neubauer et al. 1997:189)
4.1.2 “Quantitative Integration”
In 2000, Piro et al. published results from a high resolution multi-method approach to the prospection
of three archaeological sites with the aim of resolving volumetric variations in buried Roman features.
Magnetic gradiometry, resistance tomography, and ground-penetrating radar were used to survey a
range of areas between 8x8, 10x10, and 20x20 meter grids at separate sites. Piro et al. used
quantitative methods to first normalize the data sets, and then performed a variety of basic functions
on each data input to combine the survey results.
Piro outlined a normalization equation (Piro et al. 2000:204) which essentially took the absolute value
of the z value (the raw geophysical result) of the x,y coordinate of each cell, and subtracted the value
of the “undisturbed” surveyed areas. The result is then divided by the maximum raw value. This
results in producing a set of values ranging between 0 and 1 for each survey method. Piro claimed
that normalization allows for the comparison of multiple data sources, by having “deprived the data of
the physical dimensionality,” which produces new data sets that are suitable “indicators of source
occurrence” (or ISO) of archaeological anomalies. (Piro et al. 2000:204)
The first function performed on the new normalized data sets was an equation that used the sum of
the normalized values of each data set (z) and divided them by the number of methods used,
otherwise known as the average, or mean of the ISO function. (Piro et al. 2000:204) This function
indicated the spatial locations of anomalies observed with at least one or more survey methods. The
second function performed on the data was designed to detect the spatial distribution of anomalies
that were indicated by all survey methods. (Piro et al. 2000:209) This function essentially factored or
multiplied the z values between each method.
Piro observed that the first function (the average) indicated the geometry and location of several tomb
chamber features more clearly, while successfully integrating the contributions of each method into
one output. (Piro et al. 2000:209) Piro claims that this type of integration is more likely to avoid
misinterpretations of prospection data, (Piro et al. 2000:209) by creating a mechanism for the ‘checks
and balances’ of anomaly signatures.
14
4.1.3 A Synthesis of Integration Techniques: Army City
In an article published in Archaeological Prospection in 2006, Kenneth Kvamme synthesized several
integration techniques used on prospection data gathered at a World War I army camp in Kansas,
USA. (Kvamme 2006a:57) The main focus of this article was to provide a synthesis of the range of
methods available for integrated data analysis (“data fusion”), with the aim of therefore improving the
theoretical understandings of the relationships between survey types and improve subsequent
archaeological interpretations of anomalies. (Kvamme 2006a:58)
The applied survey methods included electrical resistivity, soil conductivity, ground-penetrating radar,
magnetic gradiometry, magnetic susceptibility, and aerial thermal infared over a 100 x 160 meter area
(1.6 hectares). (Kvamme 2006a:57)
Kvamme divided integration methods into 3 categories: Graphical, Discrete, and Continuous data
integrations. Graphical methods outlined within this article include such mechanisms as two
dimensional overlays, red-green-blue color composites and translucent overlays. Discrete data
sources, such as binary data were used to perform Boolean operations, as well as basic arithmetic
functions to produce “unambiguous” representations of anomaly presence or absence. (Kvamme
2006a:57) K-means cluster analysis, a spatial analysis technique which assigns values from different
data sets (in this case, cell values) to classes, or clusters based on their distance from the mean.
(Lillesand et al. 2008:570) These clusters were then used to analyze correlate anomaly locations to
further understand the interrelationships between the results of the multi-method survey. (Kvamme
2006a:70)
Furthermore, mechanisms for continuous data integration were also outlined, including: the
computation of various cell statistics like the data MAX and MIN, principal components analysis
(unsupervised classification), and binary logistic regression (supervised classification) models. These
techniques were performed on the pre-processed normalized continuous data sets to produce
imagery with “high information content,” while highlighting robust and subtle anomalies within the
geophysical results. (Kvamme 2006a:70)
Each integration technique yielded slightly different results, some more useful than others at
assessing the nature of geophysical anomalies at Army City. The continuous integration techniques
seemed to yield the most insights into the geophysical data, as they highlighted both robust and
subtle anomalies. Where certain techniques offered “visually pleasing” results, others presented
opportunities for more interpretive or predictive mechanisms for assessing anomalies. (Kvamme:
15
2006a:71) In short, Kvamme acknowledges the potential for data fusion techniques to recognize
patterns between multi-dimensional data sets that may be otherwise overlooked. (Kvamme 2006a:
70) This paper, therefore, by providing a thorough overview of the available analysis methods
commonly used in other fields of archaeological research, and their applicability to geophysical data
analysis, built the initial infrastructure for this research.
16
Chapter Five: Research ObjectivesThrough this body of research, it is the author’s intent that each one of the following objectives will be
fulfilled:
1) To compare and contrast several methods for integrating geophysical data, and evaluate each
method’s contribution to archaeological interpretations at Portus.
Multi-method, or multi-parametric survey techniques are often used on archaeological sites where little is
known about the nature of the archaeology, or where varying types of archaeological features are being
observed. (Hess 1999:157) In the case of Portus, complementary geophysical methods were chosen to
maximize the potential for interpreting archaeological features in a chronologically complex area situated
between the Claudian and Trajanic Harbors. Previous surveys which used magnetometry and resistance
tomography were complemented by an area resistivity survey and two seasons of ground-penetrating
Radar. Each geophysical method was chosen to reflect the on-going research goals at Portus, including
an aim to ascertain a more comprehensive view on the nature, dimensions and depth of archaeological
features.
The chosen range of geophysical methods should target the aims of the project, and compare with
other conventional and non-conventional methods of data recovery, including, excavation data, standing
building and micro-topographic survey, as well as mechanical augering to qualify an integrated
interpretation of results. (Hesse 1999:157) The wealth of data recovered in this area was examined
extensively as part of this research, and the results were compared and contrasted to assess the
effectiveness of each method at Portus.
2) Assess the applicability of a variety of remote sensing, statistical, and mathematical techniques for
analyzing multivariate geophysical data results at Portus.
Though the use of multiple geophysical methods in prospection is not uncommon, particularly when the
nature of the archaeology requires an integration of survey types to derive meaningful results, the
integration of results does not always accompany a multi-method survey. As indicated in Chapter 4,
recent publications in archaeo-geophysical research have provided an insightful glimpse into the
potential for the exploration of geophysical data sets through the use of techniques common to remote
sensing and statistical analysis. The inherent spatial nature of geophysical data easily allows for the
seamless integration into a GIS, allowing for a wide range of spatial analysis techniques to be applied to
17
each separate data set, with the goal of identifying patterns within anomaly locations between survey
methods.
Within this research, as in Kvamme’s (2006a), the integrated data analysis techniques were divided into
Graphical techniques, Discrete Data Analysis, and Continuous Data Analysis. The resulting output from
each mechanism was analyzed, as well as the applicability and limitations of each method within data
analysis at Portus.
3a) Provide a more holistic view of the subsurface at Portus through obtaining additional data
concerning the nature, depth, and archaeological context of subsurface anomalies, and therefore,
3b) Successfully produce a large-scale multi-method geophysical data set of an archaeological site.
Results from the variety of mechanisms for data integration produced a series of positive and negative
patterns between each data set. These correlations can then be assessed and defined, and
consequently a more holistic view of the observed archaeological feature can be ascertained. This
outcome provided a more ‘secure’ mechanism for speculating about the types of observed features,
furthering the archaeological understanding of the prospection area. A large scale integration will
provide a mechanism for identifying and analyzing the extensive features, providing an arena for an
inductive approach to additional feature identification.
4) Assist with planning for future excavations and research at Portus.
Each geophysical method that has been employed at Portus prior to this research was chosen
according to a combination of the research goals, the nature of the archaeology, and the geophysical
signatures. The extensive magnetometry survey subsequently assisted researchers in targeting an area
for excavation at Portus, and the resistance tomography survey revealed the depth of overburden in the
chosen area (De Gaetano and Strutt 2007:4), further refining research strategies prior to the
commencement of excavation in 2007. This data analysis will provide an additional platform for further
examining and targeting areas of interest for excavation and data recovery at Portus in 2009. The
reciprocal nature of research at Portus will then provide ground truth data from within the prospected
area, to allow for the continued reinterpretation of the port complex.
18
Chapter Six: Methodology6.1 Data Collection and Processing
In addition to the previous magnetometry and resistance tomography surveys, (as described in Chapter
2) the area resistivity survey and ground-penetrating radar surveys were conducted in May 2008, with
some supplemental data capture which occurred throughout the excavation season of September 2008.
Mechanical auger samples were also taken during the excavation season throughout the archaeological
park, and within the region which was originally surveyed by tomography and GPR.
Due to the extent of archaeological prospection presented within this research, each method is
presented as a separate section within Chapter 6, prior to discussing the subsequent data integration
analysis in Chapter 7. A basic understanding of geophysical methods and their use within
archaeological research are assumed.
Figure 4: Portus: Red indicates the side of the Trajanic Harbor (VI) of interest
6.1.1 Magnetometry
As mentioned in Chapter Two, the magnetometry survey data being used within this research was
collected as part of the a large-scale extensive survey of the surrounding region around the ports
complex. Only the area where overlapping geophysical surveys were conducted, to the NW of the
Trajanic harbor, Side VI, (See Figure 4) is being used in this research. The specifics of the survey
methodology are presented here, as they are reported in the Portus volume. (Keay et al. 2005)
19
The Survey
Magnetometry and it’s use within archaeological prospection are governed by the basic principles
of magnetism and rely on the ability of various instruments to detect the magnetic fields of
subsurface archaeological features. (Gaffney and Gater 2003:36) A magnetic contrast must exist
between features and their surrounding soils to detect subsurface archaeological features.
(Gaffney and Gater 2003:39)
The magnetometry survey was conducted using a GeoScan Research FM36 fluxgate
gradiometer with automatic data-logger. A 30 x 30 meter grid spacing was established using a
Total Station, and data were collected every 0.5 meter along 1 meter parallel traverses. (Keay et
al. 2005:64) The sensitivity was set to 0.1 nT (nanotesla) and the zero drift was logged at the end
of each grid. (Keay et al. 2005:64) In total, the magnetometry survey covered an area of c.178
hectares, though only a portion of data near the Palazzo Imperiale will be used for this research.
Data Processing
The magnetometry data underwent traditional processing techniques to remove shifts in the
earth’s magnetic field, to minimize signals from geology, and to enhance archaeological
responses. (Keay et al. 2005:65) The data was processed by Kristian Strutt and Julia Robinson
using Geoplot 3.0 software. (Keay et al. 2005:65) These processing steps included but were not
limited to:
• Despiked to remove high response readings from ferrous materials
• Zero Mean Traverse to average variations in the earth’s magnetic field
• Low-pass Filter to reduce the amount of variability in the gradiometer readings and smooth the
image output
6.1.2 Resistance Tomography
As mentioned previously, a resistance tomography survey in conjunction with targeted, shallow auger
sampling was conducted near the site of excavation at Portus in May-June 2007 by the
Archaeological Prospection Services of Southampton and the British School at Rome as part of the
first phase of the Portus Project. The data was gathered with the goal of determining the extent of
overburden between the Trajanic and Claudian harbors, and to inform the excavations which began
in September 2007. The full interpretations and presentation of this data is presented in the “Report
on the Geophysical Survey at Portus: May-June 2007” (De Gaetano and Strutt 2007).
20
These results were originally intended to be included as part of the integrated analysis in this
research, but due to several factors (to be addressed in Chapter 8), full data fusion with this data set
was not explored in this research. Although the results were not involved in the integration analysis,
the data heavily informed the interpretations of the GPR, resistivity, and magnetometry data, thus the
field and processing methodology are briefly discussed here.
The Survey6
Resistance tomography (or pseudosections/electrical imaging) is a archaeo-geophysical survey
technique that, like traditional resistance surveys, sends an electrical current through a series of
probes along a given traverse to detect the nature of subsurface stratigraphic changes, and
potential archaeological anomalies. This method is based on the notion that as the distance
between probes is increased, the vertical depth of detection also increases. (Aspinall et al. 1997)
“In principle, by systematic linear survey over an object of interest with increasing probe
separation, it becomes possible to assess the vertical section of the body.” (Aspinall et al. 1997)
For this resistance tomography survey, a Geoscan RM15 resistance meter with a PA3 probe
system was used. Four separate probes were arranged in an expanding Wenner array with 1.0
meter probe separation, with readings taken at the center point of the array. Eight traverses were
collected along each of the seven profiles of varying lengths. The profile names correspond to
the easting coordinate within the arbitrary excavation grid. The probe array was increased by 1.0
meter each traverse, maintaining the collection of readings every 1.0 meter. By expanding the
probe separation, readings were therefore increasing by a depth of 0.5 meter with each traverse
to build a three dimensional profile of the subsurface resistance readings.
Data Processing
The resistance readings which were recorded by hand in the field, were data entered into an
Excel spreadsheet. The readings were then converted into apparent resistivity,7 and the relative
topographic points were added for modeling in Res2DInv, the software package used to model
the resistance tomography profiles.
After the data was modeled with the topography, a least square mean inversion was applied
before displaying each section as color profiles. This has the visual effect of amplifying the
21
6 (De Gaetano and Strutt 2007:11)
7 Apparent resistivity is determined from Ohm’s law, using the potential difference (voltage) between two probes for
a known current reading. (http://www.geol.lsu.edu/Faculty/Nunn/4002_1/chp5.html Accessed 18/10/2008)
resistance readings to display greater variation within the results. The results can be seen in
Appendix B, and all in-depth interpretations can be found of individual features in the report (De
Gaetano and Strutt 2007).
6.1.3 Augering
Hand auger samples were taken in conjunction with the resistance tomography survey in May-
June 2007 to determine the depth of overburden in preparation for the excavation survey that
year. A total of 28 auger samples were taken along the resistance tomography profiles to depths
up to 3 meters, with a concentration of samples located within the excavation area. (De Gaetano
and Strutt 2007) In September 2008, 9 mechanical augers were sampled up to depths between
10-13 meters throughout the excavation area, and the archaeological park.8 These samples, in
contrast to the hand auger samples, were taken at much greater depths to assess the
archaeological, as well as the geomorphological deposits within the area. The results of both
surveys, are unpublished results, yet were used as additional comparative data set to correlate
anomaly depths in the geophysical results.
6.1.4 Resistivity
In May-June 2008, an area resistivity survey was undertaken in the areas west and south of the
excavation which began in September 2007 at Portus. The resistivity survey was conducted as part
of the Portus Project with the assistance of the British School at Rome (BSR), and the Archaeological
Prospection Services of Southampton (APSS).
The Survey
At its essence, this prospection method involves passing an electrical current through the ground
by inserting electrodes into the subsurface, and measuring the ratio of resistance to the current in
ohms(!).9 “Resistance can be established by measuring the current flowing through a body of
material and monitoring the change in voltage across the material.” (Gaffney and Gater 2003:28)
High resistance, or “positive anomalies,” can be created by features which force the current to
flow through an easier, longer path. (Clark 1990:37) Conversely, low resistance, or “negative
22
8 In collaboration with Jean-Philippe Goiran and Ferreol Salomon (Universite de Lyon) and the Portus Project
9 As measured according to Ohm's law which states that the current through a conductor between two points is
directly proportional to the potential difference (voltage) across the two points, and inversely proportional to the re-
sistance between them. (http://en.wikipedia.org/wiki/Ohm%27s_law Accessed 10/15/08)
anomalies,” lower the potential gradient while supplying easy paths for current flow. (Clark
1990:37)
A GeoScan RM15 resistance meter was used to survey using a 0.5 meter probe separation on
an 30 x 30 meter grid spacing. A multiplexer with twin probe array was used to survey two 0.5
meter transects simultaneously, doubling the rate of data collection. Grid corners were surveyed
using a Total Station, allowing for the migration of the post-processed resistivity data from
Geoplot into the GIS.
Data Processing
The resistivity data also underwent traditional processing techniques to reduce edge effects,
minimize spikes in the data, and filter signals from geological responses. The data was
processed using Geoplot 3.0 software. These steps included but were not limited to:
• Despiked to remove high responses
• Edge Match to calibrate the overall readings gathered across the survey
• High Pass Filter to remove the underlying geological anomalies
• Low Pass Filter to improve the response of ‘weak’ archaeological anomalies
• Interpolated to enhance and smooth the image for visibility purposes
6.1.5 Ground Penetrating Radar
This portion of the extensive ground-penetrating radar (GPR) survey at Portus was completed in
conjunction with the resistivity survey in May-June 2008. The portion of this ongoing survey that was
used for this research covers circa 8085 m", approximately the size of a 90 x 90 meter grid.
The Survey
Ground-penetrating radar involves “the transmission of high-frequency radar pulses from a
surface antenna into the ground.” (Conyers 2004:1) Electro-magnetic waves are generated and
released from the antenna and either attenuated, absorbed, or conducted within the subsurface
by “buried discontinuities.” (Conyers 2004:25) The time taken between when the pulses are
emitted, to when they return as reflections to the antenna, is measured to determine approximate
depth of subsurface materials. (Conyers 2004:2) The time for returned reflections is then
calibrated in nanoseconds (billionths of a second), and using velocity analysis techniques, can be
converted into depth below ground surface. (Clark 1990: 119) Timeslices 10 provide the
23
10 A series of subsurface plans at increasing depth within GPR data. (Gaffney and Gater 2003:47)
opportunity for geophysicists to analyze refections at varying depths on the horizontal plane.
(Neubauer et al 2002:142) and were used as part of the integration analysis.
The survey was completed using a Sensors and Software/Noggin plus, 500 Mhz antenna, with
an estimated ground penetration of 3.5 meters. This monostatic system11 was useful in the
Portus survey due to the cart and buggy design, and ease of maneuverability over uneasy terrain.
Traverses were collected at a 0.025 x 0.5m interval at 512 samples per scan, or trace,12 with a
setting of 4 stacks 13 in a zig-zag (forward and reverse) direction. The depth was set to 70
nanoseconds to maintain the desired ratio between the depth penetration and resolution of
recorded reflections.
Due to the nature of the landscape surrounding the excavation area, logistical complications
which arose during the collection and post-processing of the GPR data presented numerous
challenges. The existence of obstacles within the chosen survey area such as trees, fences,
exposed structural remains, and extreme elevation changes imposed constraints on the manner
in which the GPR data was collected. Unlike the Geoscan prospection instruments employed for
the magnetometry and resistance surveys, the Sensors and Software GPR does not allow for
“dummy readings” to be logged in areas where data capture is impossible. As a result, a line
survey strategy was adopted rather than the traditional grid survey typically used in large scale
GPR prospection. Collecting the data as a line survey required the meticulous recording of the
cardinal direction of each traverse, as well as the beginning and end coordinates of each line
using the established excavation coordinate system. This methodology then allowed for the
starting and stopping of multiple lines per traverse, where obstacles were present. Figure 5
illustrates each individual traverse of the total area surveyed.
24
"" System which uses a single antenna for both the transmitting and receiving radar pulses (Conyers 2004)
"! A trace is defined as a “series of reflected waves derived from one transmitted pulse or incrementally sampled
from a continuous series of closely spaced pulses.” (Conyers and Goodman 1997:67)
13 Stacking is a procedure that (with this instrument) occurs during field data acquisition which averages successive
traces to reduce interference and minimize variability in amplitude reflections. (Conyers and Goodman 1997:68)
Figure 5: Illustration of GPR east-west traverses
Data Processing
For this research, all processing of the GPR data occurred within GPR-slice imaging software,
before being exported into ArcGIS for integration analysis.
Pre-Processing for GPR-slice
Many steps were taken to assemble and prepare the GPR data for processing within
GPRslice with the aim of producing horizontal, topographically corrected timeslices to be
used in the integration analysis.
‘Sensors and Software’ exports each raw radargram from the instrument as two files. The
file containing the radar reflections has the file extension .dt1 with an accompanying header
file with the extension .hd. The header file records the specifics regarding the data collection
such as scans per trace, the length of traverse, number of stacks, etc., allowing for easy
data migration into imaging softwares such as GPR slice. However, because the data was
gathered as a series of lines instead of grids, the coordinate data recorded in the field was
25
required in addition to the header file to ensure the correct spatial arrangement of each
radargram. Consequently, the coordinates and traverse directions of each radargram were
entered into a comma delimited text file in the format of the info.dat.14
Next, all 654 radargrams and header files were imported and transferred into GPR-slice.
After the file transfer was completed, the ungained data was then converted using the
Noggin 16 to 16 bit conversion and a gain curve was applied (Figure 6) to boost and recover
deeper reflections within each radargram.15 All radargrams collected in the western direction
were then reversed.
Figure 6: Gain Conversion on Raw Radargrams
26
14 The info.dat file is the information file with which all processing within GPR slice is based. This file tells the soft-
ware the dimensions and spatial locations of each radargram, as well as concatenating the information contained
within the .hd file.
15 The conversion of radargrams is a procedure undertaken only if the equipment used in data acquisi-
tion gathers “ungained” data. Range gaining is a procedure applied to emphasize and recover deep
reflections with lower amplitudes for greater visibility. (Conyers 2004:91) This procedure can occur during
data acquisition, or post-acquisition within the chosen software package. The Sensors and Software
hardware used for this survey does not apply gain during data acquisition, thus this conversion from 16
bit “ungained” data to 16 bit “gained” data is necessary for subsequent processing steps.
Artificial markers were applied to each profile using the Sensors and Software odometer data
to ensure correct location of reflections along each traverse. This step was repeated with
each subsequent set of radargrams generated throughout the analysis of the GPR data.
Resampling and Gridding
The raw data was then resampled at the time zero point16 at 86 scans, and sliced into 25
horizontal slices. The xy coordinate data was then added to each resampled .dat file, and
gridded using Inverse Distance Weighting with a 10 cm cell size. Each time the radargrams
were filtered, the data was resampled and gridded to maintain a strict record of each
processing step, and therefore track the changes in the dimensions of anomalies in each set
of timeslices.
Filtering the Data
• Background Filter
The data exhibited some strong banding across all radargrams, as illustrated in Figure 7
(A,B) and indicated by the white and gray arrows. A background filter exists in GPR-slice
which removes horizontal banding by summing all of the reflection amplitudes recorded
at the same time, and dividing this by the number of traces. (Conyers and Goodman
1997:78) A background removal filter was applied to all resampled radargrams at a
length of 555.
• Bandpass Filter
Many lines also exhibited spikes and coupling in certain reflections, thus a bandpass filter
was applied to the background filtered data between 199 and 1000 mHz. This
successfully removed various banding and signal interruptions displayed in a few of the
radargrams indicated in Figure 7 (C) by the yellow circle.
• Migration and Velocity Analysis
Migration was performed on the bandpass filtered data to ‘collapse’ the hyperbolic
reflections caused when the antenna moves across buried cylindrical or rounded
objects. (Goodman 2008:136) One such hyberbola can be seen in Figure 7 (D), and is
indicated by the blue line. By returning the “energy along the branches of the hyperbola
to its apex,” the process of migration allows for the production of timeslices which are
27
16 The time zero point is the point at which the first recorded reflection from the ground surface is observed by the
antenna. (Conyers 2004:91)
more closely representative of the actual dimensions of buried anomalies. (Leckebusch
2003:218) An example of a migrated radargram can be seen in Figure 7 (E).
Migration requires the velocity, or the rate at which the radar wave travels through the
subsurface, to collapse hyperbolic reflections and make time-to-depth conversions.
Several reflected and direct wave methods exist for calibrating the velocity of the radar
waves. Reflected wave methods require the reflection of radar waves off of stratigraphic
sequences or features at known depths. (Conyers 2004:100) Direct wave methods
involve transmitting radar waves between antennas over known distances, or they can
involve the performance of direct laboratory tests on the relative dielectric permittivity
(RDP)17 of samples taken from the site. (Conyers 2004:100) In addition to these
traditional methods for calibrating velocity, GPR processing softwares have now begun
to incorporate the ability to calculate the velocity rate based on the length of the
‘hyperbolic branches.’ According to Leckebusch, when using this method, the velocity
calculation is correct when “the remains of the hyperbolas are removed and very fine
‘smileys’ appear.” (Leckebusch 2003:218) Using this method, the migration of this data
set calculated the velocity to be at 0.092 meters per nanosecond, with a dielectric
permittivity of 10.51. This RDP coincides with the laboratory calculations for dry silts and
sandy coastal environments. (Conyers 2004:47) The velocity calculation was then used
to convert the two-way radar wave travel times into depths below ground surface.
28
17 RDP is the “measure of the ability of a material to store a charge from an applied electro-magnetic field and then
transmit that energy.” (Conyers 2004:45)
Figure 7: Processing Steps for removing noise from the Radargrams
3D Views
The migrated radargrams were resampled and gridded using the same sample settings as
previously described. The resulting grids were then filtered using a 5x5 low pass filter, and 4
interpolations between each time slice were created totaling 76 resulting grids. Interpolation
enabled a mechanism for examining the transition of reflections between the existing
timeslices. These interpolations were then used to create a 3D volume of the timeslices,
allowing for a more comprehensive view of the data from the x, y, and z perspective. One
example of such a slice can be viewed below in Figure 8.
29
Figure 8: An example of a GPR x-axis slice showing high amplitude reflections
Correcting for Topography
The complexity of radar paths between the surface antenna and subsurface reflectors often
produces heavily skewed images of the subsurface when surveyed across topographically
complicated features. (Goodman et al. 2006:159) Reflections returning to the antenna may
not exist directly below the antenna, thus creating an offset, proportional to the topographic
changes and the tilt of the antenna over a buried feature. The extreme topographic changes
at Portus, in theory, necessitated correction for topography as well as antenna tilt.
GPR-slice contains two mechanisms for topographic correction, one which corrects the
individual radargrams for topography and antenna tilt, and one which “warps” the horizontal
timeslices to the surface topography. Initially, many attempts were made to topographically
correct the individual radargrams for tilt. The static menu was used to import an ascii file of
the modern topography, and to grid the area contained within the GPR survey. Individual
topography files (.ctm) were created for each radargram, and imported into the static menu
interface to be viewed and smoothed.
For reasons beyond the comprehension of the software creator (Dean Goodman) and all of
those involved with the processing of the GPR data, the software crashed with any attempts
to batch correct the 654 radargrams. It was determined that splitting the data into 25 meter
30
swaths allowed for batch processing and produced topo-corrected radargrams. However,
once these 25 (out of 654) profiles were resampled and timesliced, the end results seemed
to produce timeslices of only the highest amplitude reflections in each radargram. Given the
limits of the time allocation for this dissertation, the radargrams remain uncorrected.
Next, attempts were made to topo-correct the horizontal timeslices. A 3D volume (in
addition to the one created previously) which incorporated the topography was created to
warp each timeslice. However, due to the vast area covered by this survey, this topo-3D file
was almost 1GB large (nearly 10 times the size of the largest file used by the software
creator), preventing OpenGL, a memory dependent 3D volume viewer in GPRslice, from
viewing it, and causing the software to crash.
While seeking to troubleshoot this issue of topo-correction, a 32 radargram swath, East of
the path and south of the Mura Constantiniane was extracted from the data set. With the
instructions of Dean Goodman, this subset of data was used to create a new, considerably
smaller topo-3D volume. This data was then viewed in OpenGL, with topo-corrected x, y,
and z plane timeslices (Figure 9). In addition, iso-surfaces were created of the highest
amplitude reflections and overlaid on the timeslices, to reveal possible subsurface “structural
remains” pertaining to the warehouse complex south of the Mura Constantiniane (Figure 10).
The successful creation and visualization of this portion of the data in true 3D further
confirmed that one of, if not the contributing factor to the previously described difficulties
encountered during the topographic correction was the size of the data file.
Figure 9: Topo-corrected z-axis GPR timeslices of experimental swath, the top contains an isosurface
31
Figure 10: Topo-corrected z-axis GPR timeslices with isosurface render of experimental swath
Nevertheless, despite the warping success with a portion of data, the entire data set still
remains uncorrected for tilt and topography. In the end, after much deliberation, it was
decided that due to the time constraints of this dissertation, and the sheer size of data,
(despite the theoretical implications) the GPR data must remain uncorrected for topography
and tilt at this time.
Overlay Analysis
However, an additional function within GPR-slice deemed “Overlay Analysis” proved to be
beneficial in viewing the GPR data at multiple elevations below ground surface. This function
allows for the user to overlay the highest relative amplitudes from multiple time slices into one
image, producing a slice of the GPR data at multiple depths simultaneously. After extensive
manipulation of the transformation displays of the data, an overlay timeslice incorporating
amplitudes from up to 1.35 meters was exported from GPR-slice and gridded in ArcGIS (as
described in Section 6.2.1).
6.2 Data Preparation
32
6.2.1 Migration to the GIS
All data integration described in Section 6.3 was done in ESRI’s ArcGIS, thus all of the survey input
data was required as ascii data in ESRI’s grid format before analysis was performed. All data analysis
was performed on the excavation Local Grid which was established in May of 2007 during the
resistance tomography survey.
Magnetometry Data
The magnetometry data, as described previously, was gathered as part of a previous survey
season prior to the creation of the local excavation grid at Portus. The data was collected on a
Universal Transverse Mercator grid, thus all of the data was aligned in reference to true magnetic
north within Geoplot. Although Geoplot has a tool for rotating the results, the data still lacked
spatial context. The only mechanism for placing the magnetometry data in the correct spatial
context (on the local excavation grid) was to georeference the gridded data. The magnetometry
data was exported from Geoplot as a text file, and plotted in ArcGIS as xyz point data. Then the
feature data was converted to raster data using the “Feature to Raster” tool. The raster was then
georeferenced to the local grid using existing excavation data, and other geophysical data sets.
However, in ArcGIS, the georeferenced values are automatically saved as an “integer” rather than
“floating point” causing some degradation of the data. To avoid additional degradation of the
data, the ungeoreferenced data was multiplied by 10,000 using map algebra, and converted to
an integer raster prior to georeferencing it. Then, after georeferencing, the results were divided
and converted to the previous floating point values.
In addition, a low pass filter was also performed to smooth the magnetometry data, as it was
gathered at a lower resolution than the other two data sets.
Resistivity Data
Because the resistivity data was collected on the same local excavation grid at Portus, it was
exported from Geoplot as straight xyz point data in the form of a text file. This data included the
“dummy readings” which were logged in place of obstacles such as trees, fences, and standing
structures. The data was displayed as xyz points in ArcGIS and the “Feature to Raster” tool was
used to create a floating point raster, based on the z value. The “Set Null” function was
performed to transform the dummy readings (-9999) into a NoData format recognizable by
ArcGIS.
GPR Data
33
The GPR data was exported from GPR-slice in multiple ways. In terms of raster visualization
capabilities, GPR-slice contains a wide range of pre-set histogram transforms which allowed for
easy representation of archaeological features within the timeslices. Thus, for visualization and
interpretation purposes, the timeslices were exported as jpegs with world files and viewed within
ArcGIS.
In addition, the data was also exported as series of “Surfer” ascii grid files from the grid menu.
The header files of the ascii were edited to match the ESRI format for gridding, as illustrated in
Table 1.
Surfer Grid Header Format ESRI Grid Header Format
DSAA
525 770
1935 2040
4920 5074
0 113602322
1.70141000918783+38
ncols 525
nrows 770
xllcorner 1934.75
yllcorner 4919.75
cellsize 0.20
NODATA_value 1.70141000918783E+28
Table 1: List of header files for ESRI grid conversion
Due to the format at which Surfer reads/writes grids, the “Ascii to Raster” tool in ArcGIS
produced a floating point grid which was upside down, thus, the “Flip” tool was also used to
move the data to the correct spatial context. The data was then resampled and extracted to
match the resistivity data boundaries, with a 0.5 cell size.
6.2.2 Preparation for Analysis
Achieving Normalcy
After gridding the data within ArcGIS, the data was subjected to transformation in order to give
the range of values contained in each geophysical data set a normal distribution with similar
orders of magnitude. (Baxter 1994:45) This process is particularly important for such operations
as cluster analysis and principle components analysis where each band or class must have
similar ranges of values, to be given equal weight.
Using map algebra, each data set was put through a series of simple maths to reach
normalization. Each data set was manipulated to achieve a range between 0 and 1, where 0
equals the most negative values and 1 equals the most positive values in each data input.
34
Binary Data Generation
To form the input for certain Discrete data integrations, binary data was generated for each
geophysical data set. The reclassification values were obtained through visual examination of
known anomaly data ranges before generating value ranges which were representative of the
presence (1) and absence (0) of archaeological anomalies. The chosen values are represented in
Table 2:
Geophysical Method Previous Cell Values (x) Reclassified Values
Magnetometry x < 0.361821908 0
x > 0.361821908 1
Resistivity x < 0.345231605 0
x > 0.345231605 1
GPR x < 0.393026647 0
x > 0.393026647 1
Table 2: Classification values for Binary Classification
6.3 Data Analysis
“Archaeology is not about collecting data - it is about using data to understand and explain the past.”
(Van Leusen 2001:581)
6.3.1 Traditional Interpretation Methods
Several standard digital visualization methods were used to represent the geophysical anomalies in
each survey method. Though, previous presentations of the geophysics at Portus were primarily
completed using Computer Aided Design packages such as Corel Draw, the digitization and
visualization of the data within the integration analysis were completed using ArcGIS, and graphical
representations of the all 3D data was done in ArcScene and GPR-slice.
All digitization of geophysical features result from a close collaboration between the project
archaeologists and geophysicists involved with the Portus Project. Extensive interpretations of
geophysical features, and specifics regarding feature naming conventions can be found in Chapter 7,
entitled “Results.”
Digitization of Interpretations and Initial Results
35
The original interpretations from the magnetometry survey were digitized in Corel Draw and
published in the Portus volume (Keay et al. 2005), however were not originally imported into the
GIS. These were re-digitized using “head’s up digitizing” in ArcGIS, for the area in question (Side
VI of the Trajanic Basin) and used for integration analysis. The results of this survey are presented
as reported in Keay et al. (2005), and more thorough explanation of the results can be found
within.
Approximations of 2D resistance tomography features were digitized as found in the “Report on
Geophysical Survey at Portus May-June 2007” (De Gaetano and Strutt 2007).
High amplitude GPR, as well as positive and negative resistivity anomalies were also digitized.
6.3.2 Graphical Data Integration
Integration using graphical overlays and composite images is a simple and easy mechanism for
viewing separate geophysical data sets together in their spatial contexts. These techniques are often
used in archaeo-geophysics as way to visualize and interpret separate data sets, but are often
overlooked as a means for data integration. These images can be found in Appendix A, Maps
#23-26.
Two Dimensional Overlays
Several two dimensional overlays were created to visualize the geophysical anomalies within each
raw data set. Contour lines were generated for both the magnetometry and the resistivity data
and overlaid on the relative data sets. This mechanism is particularly helpful in discerning sharp
differences in geophysical signatures, and clearly defines linear archaeological features. The
contours represent the raw data intervals, thus a variety of resolutions and intervals were
experimented with to best represent each data set. It was decided that 0.75 separation for the
resistivity, 5 separation for the magnetometry, and 20 separation for the GPR data best defined
archaeological features, and avoided masking the underlying data set.
Translucent Overlays
Overlaying one to two data sets with different transparencies on top of an opaque data set
produced an additional mechanism for visualizing multiple methods. One criticism of this
technique is that the overlays often produce a “muddy” effect, masking the viewers ability to
make out which features relate to which geophysical survey method. (Kvamme 2006a:63)
Though the production of such visualizations are not grounded in any particular theoretical
36
approach, the results proved helpful in emphasizing and visualizing positive and negative
anomalies within each data set, however, easily became confusing with too many color
combinations.
RGB Color Composite
Using the Composite Tool, the three normalized data sets: magnetometry, resistivity, and GPR;
were all assigned to each of the three bands, red, green, and blue, respectively. This model was
the first multi-banded raster created from the geophysical data, and provided a simple and easy
format for manipulating and visualizing the different survey results. Though this particular
combination particularly emphasizes positive features, manipulating and inverting the band
assignments can achieve a variety of color combinations, therefore emphasizing different types of
features, positive and negative. With these band assignments, a variety of colors on the visible
spectrum revealed strong geophysical responses from multiple survey methods, including but not
limited to:
Color Geophysical Indication
Black -> Red Positive Magnetic Anomaly, other methods weak.
Black -> Green Positive Resistivity Anomaly, other methods weak.
Black -> Blue High Amplitude Electro-magnetic Anomaly (GPR), other
methods weak.
Yellow Positive Magnetic and Resistivity Anomaly
Magenta Positive Magnetic and Electro-magnetic Anomaly (GPR)
Cyan Positive Resistivity and Electro-magnetic Anomaly (GPR)
White Positive Response in All 3 Geophysical Methods
Table 3: RGB Color Interpretations
3D Integration
As described in the Chapter 5, three dimensional iso-surface models were created for a small
swath of GPR data south of the Mura Constantiane. One limitation of these graphics (Figure 10),
though a powerful means for conveying the GPR data in a visually appealing and theoretically
accurate manner, is the inability for integration analysis.
37
In addition, digitizing the GPR data, while maintaining the relative 3D attributes of each timeslice
from which the anomalies stemmed, proved a difficult task. Data was attributed according to its
relative timeslice (as described in Chapter 7), and attempts were made with various color
combinations and line widths to produce a visually appealing, yet ‘information-rich’ image for
analysis. The limitations of 2D platforms, such as a GIS, limit and at times, prevent, the true
integration of 3D data volumes such as GPR and resistance tomography data sets.
Consequently, a simple method was developed for viewing the GPR vector data in three
dimensions, using ArcGIS and basic feature class editing tools. This method involved first
extracting the surface elevations of each feature from the digital elevation model (DEM) produced
from a micro-topographic survey of the site.17 Subsurface elevations were then approximated for
each feature, based upon the corresponding timeslice depth (calculated using the velocity
analysis discussed in Chapter 5) from which the they were derived, and subtracted from the
surface elevation. The new elevations were added to the attribute table of the shapefile, and the
feature class was then converted to a ‘3d feature’ using the 3D Analyst tool. The new ‘z-
enabled’ shapefile was then added to ArcScene, and plotted using the subsurface (z) value for
integration analysis with the detailed micro-topographic, excavation, and standing building survey
of the site (Figures 11, 12).
Figure 11: 3D view of digitized 3D GPR anomalies and excavation data, colored by corresponding depth
38
Figure 12: 3D view of digitized 3D GPR anomalies with building survey data
6.3.3 Discrete Data Analysis
Data is said to be discrete if the data values are distinct, separate, and can be categorized.18
Dividing data into discrete classes with definitive boundaries, has the theoretical advantage of
removing ambiguity about the location and nature of geophysical anomalies. (Kvamme 2006a:63) In
this analysis, discrete data formed either the input and the output for the operations described in this
section.
Binary Data Analysis
Using the binary data sets whose creation was described in Section 6.2.2, a variety of logical, or
Boolean operations, and simple arithmetic operators were performed to analyze the geophysical
data. In general, as in this research, Boolean operators result in grids with cells coded as either
TRUE (1) or FALSE (0). (Wheatley and Gillings 2002:105) Boolean operators are “a class of
operations that use Boolean logic to define a selection through the actions of union, intersection,
difference, and exclusion.” (Conolly and Lake 2006) Figure 13 depicts a simple diagram which
illustrates the logic behind the Boolean Operators used within this analysis. The top diagram
39
18 Online resource: http://www.isixsigma.com/dictionary/Discrete_Data-226.htm (Accessed 110508)
depicts the grid output of a Boolean Union, while the bottom diagram represents the output of a
Boolean Intersection.
Boolean Union (Boolean OR)
A Boolean Union is said to occur when one or more corresponding cell values
are true, in which case, the cell output is 1. If all values are false, then the
output is 0. Map Algebra was used to compute the Boolean Union of the
resistivity, magnetometry, and GPR data sets. The output, due to the overall
coverage of the geophysical responses, resulted with a grid with almost 60% of
the total cells classified as TRUE. The results can be seen in Map 27.
Boolean Intersection (Boolean AND)
A Boolean Intersection, as noted in Figure 13, occurs where the positive values
intersect. In this analysis, the output is a raster with cell values of True where all
3 methods detected a geophysical event. As expected, this results in a very
small number of TRUE cells, accounting for less than 3% of the total number of
cells. Results can be seen in Map 28.
Binary Sum
A simple Binary Sum was performed using map algebra to produce a
summation of the values within each binary data set. This essentially produced
a “confidence map” (Kvamme 2006a:64) of the number of geophysical methods which
observed a single ‘event’ or anomaly. The resulting raster image displayed cell values
ranging from 0 (no event observed with any method) to 3 (event observed with 3 survey
methods).
Threshold Binary Sum
This variation in the Binary Sum produced a raster which contained binary cell values
equalling TRUE (1) only if the event was observed by at least 2 geophysical methods.
Cluster Analysis
The goal of classification investigations is to discover patterns in groupings of values within a set
of data. (Shennan 1997:220) With this aim in mind, cluster analysis was used as an unsupervised
mechanism for establishing natural spectral groupings between each band of geophysical data.
(Lillesand et al. 2008:570) Here, a variant on the commonly used K-means method for
unsupervised clustering was used called the ISODATA algorithm. (Lillesand et al. 2008:570) As
40
Figure 13: Illustration of
Boolean OR (top) and
Boolean AND (bottom)
noted by Kvamme (2006a:66), cluster analysis works well with large data sets, and allows the
user to define the number of classes anticipated within the resulting data set. This is a
“partitioning cluster technique” which divides the group of values, or attributes, into a specified
number of clusters, as defined by the user. (Conolly and Lake 2006:171) The center of each
cluster is initially determined by a random selection of “seeds” and the remaining objects are
added to the nearest cluster. As new objects are added to the clusters, the cluster centers are
recalculated. After all objects have been assigned to a cluster, the sum of squared distances (the
distance between the object and the cluster center) are calculated and provided for user
assessment of the cluster allocation. (Conolly and Lake 2006:171) This process is known as
iterative reallocation, and the number of iterations can also be defined by the user, as described
in the following Section.
It should be noted that the normalized continuous data sets (Section 6.2.2) were used as the
input to produce the following discrete classes based on the covariance matrix generated
between each geophysical data set.
Creation of Clusters
Clusters were created using the ISO Cluster Function within ArcGIS, using the normalized
resistivity, magnetometry, and GPR data sets. Three classes were specified, presuming the
location of positive, negative, and background events within the 3 bands of data. A
minimum class size of 30 was used,19 with the default number of iterations (20) and sample
interval (10). This function produced a signature file outlining the layers (each band of data
input), mean vectors (the average spectral value in each layer), and covariances (the
tendency for values to vary similarly in two bands). (Lillesand et al. 2008:550-553)
Evaluation of Clusters and Classes
The signature file was then used as the input for the creation of a dendrogram. A
dendrogram is a “tree diagram” used to visualize the distance matrix between attributes and
the groups. (Shennan 1997:222, Conolly and Lake 2006:168) Other methods, such as
Ellipse plots allow for the visualization of clusters on the appropriate axes, but were not used
in this analysis.
41
19 ArcGIS Desktop Help advises using a minimum class size at least 10 times larger than the number of layers in the
input raster bands. Because 3 bands were used, a class size of 30 was the input.
The dendrogram first defines and allocates each attribute to it’s own group, then delineates
the pair of attributes with the least distance between them and groups them. (Conolly and
Lake 2006:168) The dendrogram is useful for assessing the clusters indicated in the
signature file, prior to performing the classification. The dendrogram is contained within
Appendix C.
Classification
Next, the clusters were used to classify the remainder of the geophysical data within each
raster. In this analysis, the Maximum Likelihood Classifier was used to produce a statistical
probability that a specified pixel value belonged to a discrete cluster, or class. (Lillesand et al
2008:554) This classification required each normalized band of geophysical data and the
signature file to classify the clusters in an output raster. The reject fraction, or portion of cells
that remain unclassified,20 was set to 0.0 with the assumption that every cell classified in the
analysis should belong to one of the 3 classes in the geophysical data. Each class, or
cluster was given equal weight, and a confidence raster of the classification certainty, in
addition to the maximum likelihood classification were outputted.
The output for the maximum likelihood classification was a raster classified into 3 values
(1-3). After evaluating the probability of each pixel occurring within each class, the pixel is
assigned to the class with the highest probability, given its attribute values. (Lillesand et al.
2008:555) This grid file was then filtered using a majority filter to smooth the output and
exentuate the dominant classification. (Lillesand et al. 2008:580) The majority filter smoothes
grid data by replacing it with the half, or majority value of the neighboring cells.21
The aforementioned cluster analysis processes were repeated to emphasize 2 and 4 classes
for comparison and discussion in Chapter 7.
Class Probability
In addition to the Maximum Likelihood, a Class probability function was also performed using
the signature file. This tool outputs a multi-band raster with probability layers for each
cluster. The values contained within each band exhibit the probability (0-100) that each cell
belongs to each particular class. 22 Each class probability band corresponds to the classes
42
20 http://webhelp.esri.com/arcgisdesktop.9.3/index.cfm?TopicName-Performing_the_classification
21 http://webhelp.esri.com/arcgisdesktop/9.2/index.cfm?TopicName=How%20Majority%20Filter%20works
22 http://webhelp.esri.com/arcgisdesktop.9.3/index.cfm?TopicName-Performing_the_classification
within the maximum likelihood classification grid, 1-3. Class probability 1 seems to
correspond to negative features, as Class probability 2 seems to correspond to positive
anomalies in the magnetic and resistivity data.
6.3.4 Continuous Data Analysis
The previous sections dealt with the classification of discrete and continuous data with the aim of
producing defined classes which combined and integrated each of the geophysical data sets.
Continuous data is “information that can be measured on a continuum, or scale.”23 Unlike discrete
data, continuous data can be broken down into smaller increments and can represent any number
between the minimum and maximum values within the data set. “Continuous data are naturally richer
than categorized information, potentially enabling superior data integrations.” (Kvamme 2006:66) In
this case, the continuous data input is the real number measurements from the geophysical results.
The normalized data sets described in Section 6.2.2 formed the input raster data for all functions
performed within this section, and details of results can be found in Chapter 7.
Data Sum and Product
A variety of basic summations of a the three standardized geophysical data sets were performed
using map algebra within ArcGIS. These mathematical functions involved adding and multiplying
the cell values of each raster together to produce a raster output containing the new values.
These functions, should theoretically emphasize existing anomalies, particularly those closer to 1.
Different sum combinations were made, which seemingly emphasized different positive and
negative anomalies, making the boundaries of some more definitive than others. (Map 33 and 34)
Data Max and Min
Using the “Cell Statistics” Tool in ArcGIS, the “Maximum” value was calculated to create a raster
output of the maximum cell values contained in each input geophysical data set. The resulting
grid emphasized the positive features in each survey method, including potential structural
remains and rubble spreads. (Map 35)
In addition, the “Minimum” value was calculated to create a raster output of the minimum cell
values contained in each input band of geophysical data. This raster seems to correspond to
43
23 http://www.isixsigma.com/dictionary/Continuous_Data-96.htm
“negative” anomalies within each data set, including proposed “voids” between structural
remains. (Map 36)
Principle Components Analysis
At its essence, Principal Components Analysis (PCA) is “designed to reduce redundancy in
multispectral data.” (Lillesand et al. 2008:527) As one might expect, input variables must be
highly correlated for there to be a significant reduction in redundancy. (Shennan 1997:269-270)
The closer the original variables are correlated, the more meaningful the new bands of data will
be, and thus the more information one can retrieve from the reclassification. (Shennan 1997:270)
One might suspect that the use of PCA in the context of geophysical prospection is theoretically
applicable, particularly in cases where survey methods are highly correlated (whether positively or
negatively) such as the correlation between electrical resistivity and electrical conductivity.
(Kvamme 2006:68) However, as with Kvamme’s analysis at Army City, the overall correlation
between the input data variables, or Pearson correlation coefficient: r, remains relatively low, with
the highest value at 0.2135.
The PCA was performed using the Principal Components Tool in ArcGIS, using the normalized
results for each geophysical method as input: resistivity, magnetometry, and GPR. The
correlation coefficients were plotted on a scale of -1 to +1, where -1 equals a negative
correlation, +1 equals a positive correlation, and 0 equals the absence of correlation. (Lillesand et
al. 2008) The highest correlation coefficient was the relatively low 0.2135. This can be seen in
Appendix C which contains the PCA parameters with the covariance and correlation matrix
included. (Map 37)
44
Chapter Seven: Results7.1 Interpretation of Survey Results
Previous research results at Portus are described here in conjunction with the new survey results in order
to correlate the new interpretations with prior knowledge about the site. The magnetometry survey, in
particular, offers a larger spatial context with which to interpret the extensive features within the GPR and
resistivity surveys.
Due to the range of geophysical signatures being observed at varying depths, the decision was made to
assign new anomaly numbers to the resistivity and GPR observations. For ease of comparison, the
original magnetic feature numbers, [8.11-9.4], are used in this reporting as presented in the Portus
Volume (Keay et al. 2005), as well as in the BSR report (De Gaetano and Strutt 2007) on the resistance
tomography survey [1-55]. The resistivity features have been given anomaly numbers [R1-R14] only to
identify individual structural components of interest. The GPR data, which presented some digitization
challenges, were also given their own anomaly numbers according to the timeslice, or depth, at which
they were first observed. All cardinal directions referred to within the interpretations and within the
Figures and Maps in the Appendices, represent the direction on the excavation grid at Portus, with north
running from the Trajanic Basin towards the Claudian Basin.
The area in question has undergone a series of re-interpretations with the introduction of new survey and
excavation results; thoroughly reinforcing the reciprocal nature of the relationship between geophysics
and archaeological research at Portus. Originally, cartographic and historical resources were used to
interpret and enhance the understanding of the port complex prior to the commencement of the Tiber
Valley Project. (Keay et al. 2005:1) Two of the late 19th and early 20th century resources which were
cited by researchers within the Portus Volume are publications by Rodolfo Lanciani (1864-7, 1868) and
Lugli and Filibeck (1935). (Keay et al. 2005:1, 47-50) These resources, in particular, have given
researchers at Portus prior interpretations with which to compare and contrast the new archaeological
evidence, and continue to influence the interpretations within this research.
As mentioned previously, the research contained within this dissertation was originally focused on
gaining insights into the so-called “Palazzo Imperiale,” by complementing the previous geophysical
surveys with GPR and resistivity prospection. Four areas of interest have been identified as key to
interpreting Side VI of the Trajanic Harbor within this research. As research has progressed (though
insights into the Palazzo Imperiale are still being sought) focus has also turned towards interpreting the
45
function of the massive structure which lines this side of the hexagon [8.11-9.4], as well as determining
the full extent of the “circular wall” [8.15]. In addition, the existence of a row of exposed, vaulted
structures parallel to the hexagon, [8.1] has raised questions about the function of the massive
“warehouse” and their relationship with the Trajanic Harbor. (Keay et al. 2005:98)
7.1.1 Magnetometry
As mentioned in Chapter 3, the magnetometry surveys at Portus and the surrounding region have
covered circa 180 hectares, and resulted in a greater understanding of the structural remains of the
area. The portion of the survey used here is located on Side VI of the Trajanic Harbor, within Area 8
and 9, as described in the Portus Volume. The results can be seen here in Maps 1-2.
The first feature of interest, revealed by magnetometry was a massive structure 65 meters wide and
90 meters long, cut by the modern access path which runs north-south on the excavation grid. (Keay
et al. 2005:99) Lanciani (1868) never specifically mentioned the parallel warehouses lining the Trajanic
basin, but the cartographic representation of the port depicts individual cells identical in size and
layout, to the other sides (I, II, III, and IV) of the hexagon. (Keay et al. 2005:286) The massive structure
(divided into feature units [8.11], [8.12], [8.13], [9.1], [9.2], [9.3]) follows the alignment of the hexagon,
though to date, a clear pattern in the division of rooms and features remains difficult to interpret.
(Keay et al 2005:99) Feature unit [8.11] is divided into a series of 8 meter wide rooms, with some
divisions creating 14 x 8 meter room blocks. (Keay et al. 2005:99) Each unit within the structure
seems to have different internal organization, with some containing central court yards [8.13] which
open towards either the Claudian harbor or the Trajanic. (Keay et al. 2005:103) Lugi and Filibeck
(1935) suggested that there were two, separate, two-story buildings whose access from the Claudian
harbor side were blocked by the so-called Mura Constantiniane in the Late Antique Period. (Keay et
al. 2005:287) The standing building survey conducted in September 2008, as well as prior field
inspection, observed that the Mura Constantiniane clearly contains atleast nine blocked openings
varying between 2.3 and 6 meters wide separated by ‘brick piers’ 1 to 2 meters wide, further
confirming this observation. (Keay et al. 2005:103)
Feature [8.15], or the so-called “circular wall” was detected just north of the Mura Constantiniane and
appeared to extend into the modern path, which was unavailable for survey by magnetometry. The
external wall was measured to be 35 meters, with the inner wall measuring at 28 meters. (Keay et al.
2005:99) Though the function and evolution of this structure are still under debate, visual inspection
of its relationship with surrounding buildings initially led researchers to believe it was originally part of
the Claudian Harbor complex. (Keay et al. 2005:100) In 1868, Lanciani described a theater with a
46
square quadriporticus enclosing a garden. (Keay et al 2005:286) Though Keay et al. (2005) have
acknowledged the possibility that [8.15] is the structure Lanciani described, the location of the feature
(considerably NE of the proposed location) does not support this theory. (Keay et al. 2005:102)
However, immediately to the west of [8.15] is an enclosed central square [8.14] measuring 27 meters
across which may represent a series of columns around a court yard with potential structural remains
with differing chronologies. (Keay et al. 2005:99)
A portion of the magnetometry data collected between the massive building and the quay was
obscured by near-surface rubble and noise, and a strip of 30 meters lining the quay was unavailable
for survey due to the existence of a modern day path and extensive vegetation. (Keay et al 2005:99,
103)
7.1.2 Resistance Tomography
A detailed account of the resistance tomography and auger survey results can be found in the BSR’s
“Report on the Geophysical Survey at Portus May-June 2007” (De Gaetano and Strutt 2007). Every
feature detected in the seven resistance tomography profiles will not be reported here, however, it is
necessary to speak generally about the nature of the anomalies, in order to compare and contrast the
spatial location and identification of features within the other prospection results. Individual figures
containing the results of each profile are located in Appendix B, and the results as they correlate to
other methods can be seen in Maps 3-4.
The resistance tomography was successful in identifying several features of interest on Side VI
between the Claudian Harbor and the Trajanic Basin. Hypotheses concerning the continuation of
magnetic feature [8.1], a complex of small vaulted ceiling structures lining the hexagon, were
supported with a series of high resistance readings ([1], [2], [18], [29], [44] and [51]) along the
southern ends of each tomography profile. (De Gaetano and Strutt 2007:17)
It was also observed that the archaeological deposits remain at a greater depth west of the main
access path, relative to the shallower deposits to the east. (De Gaetano and Strutt 2007:17) The
auger samples confirmed these readings, as structural foundations were reached at greater depths.
This observation has led researchers to believe that the modern topography on this side of the
modern access path is at the first floor ceiling level of the remaining structures in this area. (De
Gaetano and Strutt 2007:17)
In short, the survey was successful in estimating a varying depth of material (from 0.1-1 meter)
overlying the structural remains throughout most of the prospection area. (De Gaetano and Strutt
47
2007:19) In the area to the south of the Mura Constantiniane, it was observed that 1 meter of
deposits rest on top of a series of low resistance readings, correlating to the potential room blocks
observed in the magnetometry and resistivity data sets. (De Gaetano and Strutt 2007:19)
7.1.3 Resistivity
The resistivity results were somewhat perplexing, and slightly unexpected, particularly in the vicinity of
magnetic features [8.11], [8.12], and [8.14]. However, that is not to say that the resistivity was
unsuccessful, as these results, potentially give the opportunity for a range of interpretations. It should
be noted that vegetation presented many disturbances to the data, as illustrated when the locations
of trees were plotted over the resistivity results (Map 5). As with all of the geophysical results,
individual anomaly interpretations here have been restricted to only potential features of interest (Map
6) that may contribute to the wider interpretation of the site as a whole.
The resistivity identified multiple linear features running north-south between the Palazzo Imperiale
and the Mura Constantiniane and Trajanic Harbor. However, in the area of [8.11] and [8.12], with the
exception of [R2], low resistance features seem to match the locations of positive magnetic features,
originally interpreted to be structural features, primarily walls. A low resistant, linear, north-south
feature [R1] circa 55 meters long and 3 meters wide, was observed matching the location of a
positive magnetic wall feature dividing [8.12] and [8.13]. This feature [R1] seems to be faintly divided
by resistant east-west divisions every 8 meters, which, in fact match the room dimensions originally
observed in the magnetic data in [8.11] and [8.12]. It is possible that this feature [R1] actually
represents areas of the structure that have collapsed to reveal the low resistant sediment within the
structure, and the high resistant areas to the west are in fact the remains of the preserved ceiling.
This hypothesis might be confirmed by the observation by the resistance tomography survey that the
modern topography in the area of this feature is at ceiling level of the first floor of the building complex
associated with the Palazzo Imperiale, as referenced previously.
In addition, low resistant anomalies [R3] and [R5] are located precisely where the positive magnetic
wall that forms the most eastern boundary and northern edge of [8.14] is located. Several
explanations could be given for this observation. One potential explanation could be that low
resistance readings might be the result of the collapse of structural remains causing the subsequent
siltation of sediments, against the positive magnetic features. Or, the magnetic observation could
potentially be interpreted as a build up of magnetic sediment disturbance instead of a wall, though
this is not likely given the relative magnetic responses of the fired brick structural remains.
Potentially, different types of building material were used on this structure creating a difference in the
48
magnetic signatures in this area. Suggestions have also been made that the low resistance feature
[R3] running north-south along the eastern edge of the structure represents an “aqua-duct” or
drainage feature along the side of the structure [8.14].
Another possible interpretation of this feature [8.14] might be that the high resistant east-west
northern boundary [R4], might represent the continuation of one of the walls which formed the
Palazzo Imperiale. This possibility is correlated in the northern most section of GPR survey data with
the observation of high amplitude anomaly [G1], located between the bisected building, on top of the
modern day access path.
Contrary to the trend around [8.11], [8.12], and [8.14] where low resistance anomalies match positive
magnetic features, [8.13] displays the opposite results. In [8.13] and [9.1] several high resistant
anomalies seem to match positive magnetic structural features. [R10] and [R12], high resistant
features, match the locations of north-south walls detected in the magnetic data in [8.13] and [9.1].
[R7], a north-south high resistant feature located between two walls, where a highly magnetic feature,
10 meters in length exists. This feature has been ‘ground-truthed’ to be a brick face wall, as portions
of brick remain exposed from the face of the mound.
After comparing the locations of high and low resistance anomalies to the location of anomalies
within the resistance tomography data, there’s no surprise that there seems to be a successful
correlation between the two data sets. Map 4 illustrates the digitized anomalies which correspond to
features within the resistance tomography data, totaling at least 25 distinct, corresponding features.
Correlating and associating the two data sets allows for the depth estimations for the features from
the area resistance survey, and vice versa for the tomography data. Upon further investigation of the
resistivity data, it is noticeable that some low resistant features [31], [32], and [40] potentially shielded
high resistant features that are apparent in the resistance tomography data. This may be explained
by the difference in observation depth between the resistivity prospection (circa 0.5 meter) and the
resistance tomography (up to 4 meters).
7.1.4 GPR
As mentioned previously, the GPR results presented a number of challenges for interpretation and
digitization. With the presence of near surface rubble, as observed in the magnetometry data, and
vegetation disturbances, it was often difficult to differentiate collapse and random noise from intact
archaeological features. It was decided that the easiest way to represent the 3D volume of features
within the GIS was to digitize the anomaly locations directly from the individual timeslices. A system
49
of numbering was established which comprised of a capital letter, which represents the timeslice from
which it was initially observed, and an anomaly number. Only the timeslices which exhibited features
of interest were assigned feature numbers and included in Appendix A containing the maps (Maps
7-20). Though all anomalies were digitized, and can be seen in the 25 individual timeslices on the
accompanying CD, only major features of interest are discussed here, as they relate to features within
the magnetometry and resistivity, and only the timeslices discussed here were included in Appendix.A
The features are referred to individually in order of the depth at which they appear.
The anomalies are reported here according to their depth below the modern day ground surface, as
calculated by the velocity conversions and migration (described in Chapter 6) within GPR slice. A
critique of this measurement has been included in Chapter 8, however, for now, these depths should
be viewed as an estimation.
[A1] is a high amplitude east-west feature which first appears at 0-22 cm below the modern day
ground surface and extends west from the modern access path and cuts south, forming a right
angle. This feature seems to correspond to high resistant readings in the resistivity data, and a
positive magnetic response in the magnetometry.
[A2], [A3], [A4], and [A6] are all high amplitude, north-south alignments in varying lengths, from 5
to 40 meters throughout the 0-3 meter prospection depth. [A2] and [A3] correspond to positive
north-south alignments in magnetic feature [8.13] and [9.1]. East-west alignments extending
west from [A2], first observed at 50 cm, seem to form 2-3, 8-10 meter wide room blocks which
are present to a depth of about 1 meter. [A6] correlates to the middle dividing wall between
magnetic features [8.11] and [8.12], and [A5] seems to represent the sporadic 8 meter dividing
walls in [8.11], where high resistant features have also been observed.
[B1] is first observed at 13-34 cm below ground surface, and may represent a continuation of
magnetic feature [8.1]. When the standing building survey data was mirrored for the exposed,
above ground features of [8.1] was mirrored and moved to the area of [B1], the high amplitude
edges, or “walls” of this feature seem to match the same layout. At 50-72 cm the entire feature
is filled with high amplitude readings, which may be interpreted to represent the floor, or base of
the structure itself. The feature then dissipates around 88-110 cm, but seems to maintain an
alignment with [A6].
[E1] and [E2] are high amplitude linear anomalies which appear at 50-72 cm that form the
boundaries of magnetic feature [8.14] and low resistant [R2] and [R4]. [E1] is the north-south
50
eastern boundary of [8.14], running 30 meters in length, and coinciding with [R3]. At about 1
meter below ground surface, the high amplitude, parallel feature to the east appears, coinciding
with a negative magnetic feature. [E2] is the east-west high amplitude edge that appears on the
alignment of the Mura Constantiniane, with a low amplitude area appearing to the south. This
feature continues until around 1 meter below ground surface.
[E3] is an east-west anomaly with curvature in the northern portion of the GPR results, between
the bisected Palazzo Imperiale along the modern access path. This feature appears at 50-72 cm
and continues to around 1 meter below ground surface and could potentially represent the edge
of the Claudian basin.
[G1] is an east-west, high amplitude feature within the modern access path, which may be a
potential continuation of a wall feature which originally connected the Palazzo Imperiale. This
feature is first observed at 75-97 cm and dissipates around 88-110 cm below ground surface,
and may correspond to high resistant anomaly [R4].
[G2] is a high amplitude anomaly with a curved edge, located on the modern access path to the
east of magnetic feature [8.14] that may represent the internal and external walls of the “circular
building,” magnetic feature [8.15]. It first appears at 75-97 cm, and seems to continue through
until 126-147 cm below ground surface. The excavation survey data for feature [8.15] was
“mirrored” to estimate the presumed location of the missing portion of the internal and external
walls under the path. The anomaly, both the inner and external walls, almost precisely match the
presumed location, confirming the location of the missing portion of this feature.
[H2] is an east-west, high amplitude anomaly, west of the path, that seems to be on alignment
with the supposed southern boundary of the massive warehouse. This feature, which first
appears at 88-110 cm, corresponds to the southern boundary of magnetic feature [8.12], and
extends to 113-135 cm below ground surface.
[H3] is an east-west linear alignment that matches the trajectory for the front, southern wall of the
feature [8.1], which appears at 88-110 cm. This feature continues to 3 meters below ground
surface with extensive parallel walls which progress north with the increase in depth.
[K1] is composed of a series of high amplitude responses south of the warehouse, and west of
the modern access path. This feature is on alignment with magnetic feature [8.1], as well as high
amplitude anomaly [B1]. This may represent the continuation of [8.1], however elevation
51
differences have been observed between the depths of [B1] and [K1]. The feature first appears
at 126-147 cm, and takes many forms until depicting a clear resemblance to the structural layout
of anomaly [B1] at 214-235 cm.
[M2] is a north-south high amplitude anomaly located in the middle of the modern access path
which bisects the warehouse structure. This anomaly first appears at 151-172 cm, however
appears to be a corner, which breaks to the west on the same east-west alignment of anomaly
[H2]. This could potentially represent the southern boundary of another north-south alignment
which has been destroyed by the creation of the path.
7.2 Results of Integrated Data Analysis
7.2.1 Graphical Overlays
Two Dimensional Overlays
The two dimensional overlays provided a mechanism for viewing the significant changes in values
of one method overlaid on another. This analysis limited the 2D overlays to the use of two layers,
as the addition of the remaining data set detracted from the effectiveness of the representation of
features. This limitation makes it difficult for full integration to occur, as the user is limited to only
two out of three data to relate. In this sense, it should be stressed that two dimensional
graphical overlays. like the ones completed for this research, do not actually generate new data,
they merely give you a mechanism for visualizing the overlapping data sets in one image.
Translucent Overlays
The translucent overlays were the most simplistic data “combination,” (second only to the
contours) and as seen in Maps 24-25), provided an interesting means for viewing the 3
geophysical data sets. However, at the onset, it proved somewhat difficult to choose successful
color combinations, and secondly, interpret the resulting image. Nevertheless, the color
combinations in Map 25 created a red-orange output for positive resistivity and GPR features,
while emphasizing dark green for positive magnetic features. This image overlay was particularly
insightful in areas where positive features detected by one method, overlapped with negative
features of another method, for example at the location of the southern boundary of positive
magnetic feature [8.14].
52
RGB Color Composite
The RBG model (Map 26) was potentially the most effective in utilizing all aspects of each
geophysical data set, and fusing, them in a meaningful way. Through manipulation of the band
assignments, the RGB image proved to be a simple mechanism for interpreting the positive and
negative features, particularly in the area of features [8.14], [R3], [E1], and [E2], west of the
access path, where it was challenging to assess the precise feature boundaries using the two
dimensional overlays. The RGB composite emphasized robust features which were observed in
all methods, such as [A2] and [R14], as well as allowed for the visualization of more subtle
features that might otherwise go undetected such as the east-west resistance features across
[R1]. The RGB composite was the most effective data set produced in this analysis, and given
its theoretical grounding within remote sensing techniques, there is plenty of space for further
exploration of this data set.
3D Integration
Due to time constraints, the full capabilities of this 3D vector data set were not realized by the
completion of this research. The benefits of visualizing the 3D GPR shapes within their
subsurface locations in relation to the excavation and topographic data are apparent, however,
the strength of this method may be in the potential for a platform which also facilitates interactive
querying of the results. Figure 14 is an example of the product of a definition query for all
features below the suspected base of the Mura Constantiniane. The selection and display of
only features at corresponding depths of “key horizons” at Portus, could potentially facilitate a
clearer integration of survey methods, as well as clearer understandings interpretations of the
chronological sequence of structures in this area.
53
Figure 14: Query of all features at, or below the depth of the base of the Mura Constantiniane
7.2.2 Discrete Data Analysis
Binary Analysis
The Boolean OR function was an easy method for quickly visualizing the locations of cells where
a positive anomaly was detected. The overall spread of ‘TRUE’ values in the output was
extensive, making it difficult to delineate individual features, with the exception of feature [8.14]
where the boundary walls have been heavily emphasized with this function. (Map 27) The main
utility of this output is its ability to easily convey the locations of all positive anomalies.
The output of the Boolean AND function was a quick and easy means for visualizing the location
of anomalies, as in the Boolean OR, however in this case, produced an output with only those
which were detected by all three methods. (Map 28) As one might expect with using only three
input data sets that measure different geophysical elements, this function produced a binary
output with very limited analysis capabilities. The results convey very little about the nature of the
geophysical anomalies, as the only observation that can be made is “presence” or “absence” of
positive anomalies in all methods.
The Binary sum was helpful in ascertaining a simple “confidence map” (Kvamme 2006a) of the
presence of positive anomalies. (Map 29) This output of the Portus data sets produces an
54
interpretable map which researchers can use to assert some degree of ‘objectivity’ when making
interpretation of anomalies, yet, still only verifies existence of detection by ‘x’ methods, leaving
the viewer with the task of relating the image back to the original individual results.
Cluster Analysis
Due to the nature of the geophysical data, the appropriate or optimal number of classes to
assign the cluster analysis may not be known. (Kvamme 2006:66) Consequently, cluster analysis
was performed with a series of parameters, using 2, 3, and 4 classes. The maps for these
results can be found in Appendix A, Map 31-32. The first cluster analysis was performed using a
setting of 2 classes, intended to represent anomaly “presence” or “absence.” The filtered output
(See Section 6.3.3 for an explanation of the filter) produced a classification that corresponded to
interpreted high amplitude GPR features, and to a lesser extent positive magnetic and resistance
features (2), while class (1) corresponded to negative anomalies and ‘background data.’ The
cluster analysis was then performed with a setting of 3 classes, with the intentions of
representing positive, negative, and background data. The 3 class analysis produced a
classification that corresponded to more ‘robust’ positive features (i.e. features which were
detected by 2-3 methods) (3), positive magnetic features (2) that do not correspond to anomalies
detected by other methods, and negative features with background data as (1). Lastly, the
cluster analysis was performed using 4 classes, as an attempt to successfully extract and classify
the negative features from the background data. The 4 class analysis again created a
classification corresponding to the robust features detected by all methods (4), with classes (3)
and (2) corresponding to progressively more subtle positive features, and (1) corresponding to
negative features and background data.
After examining the class distribution, one possibility is that background data doesn’t actually
exist at Portus given the spread of structural remains, and the classification doesn’t represent
“presence” or “absence,” but rather it exhibits a “positive” or “negative” class-to-feature type
correspondence.
However, the inability of the cluster analysis to readily identify negative anomalies, might stem
from limitations of the ‘arbitrary’ unsupervised classification of cell values. This possibility is
explored in more depth in Chapter 8.
7.2.3 Continuous Data Analysis
55
Data Functions
As one might expect, the Data Sum output (Map 33) emphasized robust anomalies, yet also
included more subtle positive anomalies that were not particularly apparent in the previous data
outputs. The lowest cell value in the output of the Data Sum was 0.421, leading one to assume
the absence of a strong correlation between negative features in all 3 data sets (presumably, if
there was, the lowest value would be closer to 0). The Data Product was particularly useful for
emphasizing and exaggerating robust anomaly boundaries, and masking subtle ones. If any of
the three data inputs contained 0 values, or negative anomalies, the output cell value always
equaled 0, further emphasizing negative features in the output.
The Data Max function highlighted the most robust anomalies in each data set by taking the
maximum value from each grid. Within the output, (Map 35) all cells which equal 1, indicate the
presence of a positive anomaly in at least one of the input data sets. This result emphasized
robust positive anomalies more than any other data functions performed, as one might presume.
The Data Min function, conversely, accentuated the most negative features within each data set.
(Map 36) The output is particularly ‘spikey’ with low values, and after further examination into the
source of the spikes it was determined that most of them originated in the GPR data. Apart from
the visually unpleasant nature of the image caused by the spikes, the output is, from a
geophysical point of view, interesting. This is one of the first functions performed on the data that
has resulted in an output which has examined the negative anomalies within the geophysical data
sets.
Principal Components Analysis
As mentioned in Chapter 6, the highest correlation coefficient for the PCA output was 0.21355,
indicating a low correlation, as well as high variation between the distribution of values contained
within the three input data sets. (Map 37)
The primary purpose of this analysis technique is to reduce redundancy and produce a principal
component with higher contrast between components than in the original data sets. The
applicability of the Portus geophysical results in this type of analysis is questioned, as an
examination of the scatter plots of each method does not indicate extensive overlap between the
normalized values. As a result, the 1st principal component contains minimal contrast, and the
2nd and 3rd components are the input variables, resistivity and magnetometry, respectively.
56
Chapter Eight: Discussion8.1 Some Potential Implications of the Interpretations
Following the presentation of results in the previous chapter, it is now necessary to discuss the overall
repercussions of the individual anomaly and feature observations on the understanding of this side of the
port complex.
Assessing the nature of the geophysical signatures within the area of the warehouse on Side VI can
potentially inform the understanding of the function of this massive structure, as well as this side of the
Trajanic Basin. The structural layout of the individual cells within Features [8.11]-[9.3], therefore
potentially shapes the overall understanding of the role of the storage facility, and the role of Portus in
the transportation of goods throughout the Roman Empire. The existence of central courtyards within
the horrea at Ostia, for example, is characteristic of a venue where privatized, entrepreneurial activities
take place. (Keay et al 2005:310) Thus, the existence of warehouses with layouts that include “corridors”
rather than central courtyards supports two potential notions about the function of the port complex.
The absence of central courtyards potentially reinforces the theory that the regulation of the
transportation of goods was under more state control at Portus, relative to Ostia, as courtyards would
not be required to barter and negotiate the sale of goods. (Keay et al. 2005:310) The second possible
implication is that Portus acted as a “food reservoir” for the long term storage of goods to be shipped to
Rome, as needed. (Keay et al. 2005:310)
With that being said, Side VI of the Trajanic Basin (the focus of this research) is the only side within the
port complex where “corridor” warehouses are not the dominant type of configuration. (Keay et al.
2005:310) The implication of this observation on the theories explained above, is not yet known,
however, the warehouse layout may potentially be accounted for by the complexity of the Palazzo
Imperiale and it’s relationship with the surrounding structures. The magnetometry data, though difficult
to interpret in areas, is helpful in illustrating the layout of the warehouse features. The resistivity also
supports the location of large north-south walls, or “bays,” indicated in both the GPR results and the
magnetometry.
Upon close examination of the detailed building survey data from 2008, the location of these walls is
further supported by the existence of large concrete “pylons” at various intervals along the Mura
Constantiniane. Many projections and speculations have been made about the previous existence of
additional pylons on the same alignment, east and west of the access path, as illustrated in Figure 15.
57
Positive magnetic features on this alignment might represent the continuation of this pattern, in the
western portion of Area 8. The estimated pylon alignments also match north-south structural feature
alignments present in all three methods, as well as significant topographic changes in this area.
Figure 15: Schema of Pylon Pattern
In addition, a distinct edge between high and low amplitudes, beginning at approximately 1 meter in the
GPR data was observed east of the modern access path, south and parallel to the Mura
Constantiniane. One interpretation of this distinct edge has been to attribute it to the construction of the
Trajanic basin, yet this interpretation is under debate, for reasons evaluated in Section 8.2.1.
8.2 Complications and Critique of Methodology
8.2.1 GPR
Velocity Analysis
As explained in Chapter 6, velocity analysis of radar waves can be conducted in a variety of
ways. For this study, the velocity was calculated using hyperbolic fitting of point source location
reflections within GPR-slice. This method, though theoretically sound (Leckebusch 2003), does
not account for the local conditions of the prospection which are present at the time of the
survey.
58
Five targeted GPR grids were surveyed within the excavation in September 2008 (3 months after
the first phase of GPR survey) with the intention of having precise radar reflections to correlate
recovered feature data within trenches and sections. With minimal assessment of this data due
to time constraints, the radar reflections (with time to depth conversion) seem to correlate with
observed features. However, this research is hesitant of making a direct correlation between the
depth calculations within the excavation to those used in the areas discussed in this dissertation.
As cited by an example of the use of GPR within an excavation at Petra, Conyers warns against
the conversion of velocities based upon an excavation face left exposed to the elements for a
significant length of time. (Conyers et al. 2002) The deeper materials and sediments retain natural
moistures, allowing for radar waves to travel at lower velocities, creating an image of features that
are located at much shallower depths than observed. (Conyers et al. 2002) Though this data will
be further examined in future work, the time to depth conversions for the larger research area
remain estimates, and should be used with caution.
Correction for Tilt and Topography
As mentioned in Chapter 6, correcting for topography and antenna tilt is essential to achieving an
accurate representation of the subsurface through GPR prospection, particularly if dramatic
changes in elevation are present within the survey area. “Reflection trace shift,” is dependent
upon the velocity of the radar wave throughout the subsurface, as well as the distance between
the antenna and the reflected surface. (Goodman et al. 2006:163) As the average velocity of the
ground increases, the potential for “trace crossover” also increases, leading to even greater
distortion in the resulting reflections (Figure 16). (Goodman et al. 2006:160) Therefore, when
radargrams are not corrected for antenna tilt or topography, the location of subsurface anomalies
can shift dramatically from their actual location in the subsurface. In reference to the integration
analysis used in this research, this shift and subsequent distortion of the location and shape of
the anomaly, could potentially have an detrimental effect on the comparison of anomaly locations
between methods. This could, in theory, create positive or negative correlations or relationships
between detected anomalies, that may not exist if a more accurate representation of the
subsurface was obtained prior to the integration analysis.
59
Figure 16: The effects of slow and fast velocities (Goodman et al 2006:160)
Edge Effects
Potential edge effects are also visible in several of the timeslices, which may have created
misrepresentations of amplitude signatures, resulting in false interpretations of anomalies. For
instance beginning in timeslice [A], (Map 7) at the southern edge of GPR feature [A4], a corner
which crosses the modern access path and aligns with the western edge of the survey grid has
potentially produced an over-emphasis on the interpretation of the anomaly (a prominent feature
which extends up to 3 meters below the ground surface).
In addition, starting at approximately 1 meter below ground surface, an east-west edge occurs
west of the access path at a northing of 4975 on the excavation grid. This edge, which
continues through 3.15 meters (Map 20), has been interpreted by some to potentially represent a
very early phase in the development of the port, potentially representing the southern edge of a
channel or a constructed platform for the foundation of the massive warehouse structure. These
are all speculations, of course, however there is some concern that these high amplitude
reflections stem from elsewhere. It should be made known that there was rainfall mid-way
through the survey, and as demonstrated by extensive experimentation on the potential
contributions to radarwave variability, (Kvamme 2008) a significant shift in the moisture content of
the soil could potentially change the observed reflections.
8.2.2 Classification
Binary Data Classification
“The quality of the training process determines the success of the classification stage, and therefore, the
value of the information generated form the entire classification effort.” (Lillesand et al 2008:557)
60
Clearly Lillesand was referring to the act of determining training data sets for use in a supervised
classification, nevertheless, the same point may be made about the selection of anomaly
thresholds for the binary data classification. These thresholds, though based on a cautious
examination of the range of anomaly values within each data set, were a subjective selection of
values based on inductive reasoning and knowledge of the results. The ‘goodness of fit’ of the
chosen anomaly ranges will never be determined unless extensive ground truthing of every
anomaly takes place, which in turn, defeats the purpose of the non-invasive, inductive nature of
geophysical prospection. In short, any critique of the binary data classification is a product of the
uncertainty that underlies prospection as a whole, and as mentioned in Section 3.1.2, even under
the most ideal survey conditions, there is never a 100% certainty in geophysical prospection.
To Be, Or Not To Be Supervised
All multivariate classifications performed in this research, (cluster analysis and principal
components analysis) are unsupervised classifications which result from algorithms that “examine
the unknown pixels in an image and aggregate them into a number of classes based on natural
groupings.” (Lillesand et al. 2008:569) One critique of this method, though clearly useful for
recognizing patterns which may not be readily apparent in a data set, is that the output of such
classifications may emphasize or understate relationships between data values that may not be
useful for their applications in the relative research. In contrast, supervised classifications require
the user to define ‘useful information categories’ to be compared to the spectral signatures of
other cells within the data set. (Lillesand et al 2008:569) Where in unsupervised approaches,
results should be compared and contrasted with real data distributions, supervised classes allow
for the immediate association of results based on initial training categories. However, a critique of
supervised classifications may be made of the inherent bias engrained within the data output, as
defined by the training process. In the end, it is no doubt ideal to utilize both strategies for
determining patterns in one’s data, as both classification types act as complementary analysis
techniques, where the limitations of one are compensated by the strength of the other.
8.3 An Assessment of Limitations and Applicability
8.3.1 Discrete Data Input
The data analyses which used the binary data as input variables (including the Boolean calculations
and mathematical functions) produced the weakest output, in terms of the level of meaningful
interpretations which could be made from them. The outputs failed to convey any information about
61
the nature of the anomalies, and only indicated presence, absence, and the number of methods
which detected an anomaly at a particular spatial location. Caution should be taken when examining
these data outputs, merely because four methods observe an anomaly, does not necessarily indicate
a feature of interest, particularly when the classification of the initial thresholds was the result of a
subjective, rather than objective, means of choosing the data ranges.
8.3.2 The Number of Data Inputs
A potential limitation of the more sophisticated methods of cluster analysis and principal components
analysis techniques may be the number of input variables required to create a meaningful output.
The original research proposal included an additional field season of resistance tomography, to
increase the resolution between existing data profiles with the intent of creating horizontal
interpolations. This data set was intended to be the fourth data input. However, the nature of the
dehydrated, rainless sediments at Portus in September prevented full contact of the electrical current
and the completion of the survey.
8.3.3 The Level of Detail
A major distinction between the case study examples described in Chapter 4 and the analysis
completed for this research is the difference in the level of assumptions that can be made about the
analysis results. With recent historic archaeological sites such as Army City, researchers have the
benefit of historic records, including plans and photographs, and even oral accounts of the nature of
the subsurface features being prospected. Though antiquarians have conducted extensive research
at Portus for some time, many questions regarding the chronological sequence of the port, as well as
its relationship with other ports in Italy and elsewhere are still under debate. (Millet et al. 2004:222, as
cited by De Gaetano and Strutt 2007:6) Establishing a chronological sequence and overall plan of the
structures, including the Palazzo Imperiale, the “circular wall,” and the “warehouses” have proved to
be a challenging, and continuous forum for archaeological dialogue. Though the geophysical results
have made a tremendous contribution to the discussion about the nature of the structures at Portus,
a certain level of uncertainty still remains about the nature of the anomalies. Much of this may be
attributed to the state of remains within the area in question. As stated previously, the portion of the
site being investigated here is inundated with a significant amount of collapse and overburden
shielding the archaeology, and making interpretations of the geophysical anomalies, in terms of the
level of detail indicated by the results, difficult. The prospect of determining “four types of
floors” (Neubauer and Eder-Hinterleitner 1997:185) remains unlikely any time soon. However, in this
case a successful data fusion is not judged on the basis of one’s ability to discern the minute details
62
of archaeological features; those are merely byproducts of a series of optimal conditions which allow
for exciting, innovative finds. Here, the author has chosen to focus on the mere creation of a type of
data fusion that champions exploratory data analysis, and emphasizes positive and negative
correlation of feature existence.
63
Chapter Nine: Future Prospects & Conclusions
Proceeding the end of this research, continued analysis and processing of the prospection and
excavation data recovered at Portus in 2008, along with extensive comparisons with the mechanical
auger data, and the building survey results will occur. The extensive GPR data set, which continues
eastward from the data used in this analysis will also be processed, analyzed and interpreted. In
addition, the geophysical data results of this research will almost certainly play a role in targeting future
excavations for 2009.
With the acquisition of an automatic, multi-probe, resistance tomography kit, resistance tomography will
be used to continue mapping Side VI of the Trajanic Basin, to complement the extensive GPR done in
this area in September of 2008. This additional data source will provide complementary three
dimensional data to be incorporated into the data fusion methods described here.
Future prospects for the use of data fusion, in general, most certainly include the incorporation of the
third dimension in data analysis techniques. The three dimensional vector data created for the GPR
data provides an accessible interface for visualizing and interpreting the relationships between the GPR
results and the excavation data. The addition of the resistance tomography data, as well as models of
the standing building survey will greatly increase the researcher’s ability to correlate and interpret the
features of interest based upon their elevations. Though, elsewhere, alternate softwares have also been
used (Watters 2006) such as Amira,24 to visualize three dimensional geophysical data sets in their
context, the strength of the 3D vector data created for this research, lies in it’s simplicity. This shapefile
can be imported/exported to any 3D viewer or drawing package for interpretation, where as using
expensive proprietary softwares, often limits the full realization of the data’s potential.
New data fusion softwares are in production which import, process, analyze, and essentially fuse
geophysical data within a single user interface.25 These types of interfaces will not only encourage the
increased use of data fusion techniques but will also, in the author’s opinion, increase the level of
meaningful, progressive research within geophysical prospection, as well as permit an opportunity for a
wider understanding of the archaeology in question.
64
24 Amira is a three dimensional imaging software originally developed for the medical field. (Watters 2006: 285)
25 The University of Arkansas’ Center for Advanced Spatial Technologies: Geophysical Data Analysis Toolkit
http://www.cast.uark.edu/home/research/geophysics/geophysical-data-analysis-toolkit.html
Though the interpretations of the research conducted here have not yet been fully realized, it is the
author’s belief that the types of methodologies which were used have provided a much more holistic
view of the subsurface anomalies at Portus. The combination of integrated survey methodologies and
integrated data analysis has provided a wealth of different types of data, including resources with both
analysis and visualization capabilities, increasing the potential for future interpretations of archaeological
and geophysical features at Portus. Though each method used in this research contained strengths and
weaknesses, of all of the analysis methods used, the RBG model, cluster analysis, and 3D vector
exploration have been the most insightful, and visually pleasing results of this analysis.
The process of archaeological data integration, in general, is a process that is comprised of multiple
phases, including data collection, data analysis, and interpretation. A perpetual cycle of reevaluation is
required as new data is gathered, analyzed, or interpreted, ideally forming a continuous progression
towards a better understanding of the archaeology. Portus is no different, in that each phase of
research, from classical texts to excavation, through to geophysical prospection, is never complete, and
as new data sets are acquired additional groundwork is laid to interpret and reinterpret the history of the
port complex.
Despite the limitations of individual methods performed in the integration data analysis, it is strongly
believed that the results of the foregoing methodology have considerably increased the potential for
using geophysical prospection as a means for understanding the uncertainties inherent to archaeological
and geophysical research. The archaeological interpretations of the integration data analysis has by no
means provided a comprehensive list of conclusions, but rather provided the framework for the
continued discussion, analysis, and interpretation.
65
BibliographyAspinall, A. and Crummett, J. G. 1997. ‘The Electrical Pseudo-section.’ Archaeological Prospection Vol
4 pp 37-47
Baxter, M. J. 1994. Exploratory Multivariate Analysis in Archaeology. Edinburgh: Edinburgh University
Press
Baxter, M. J. 2001. ‘Multivariate Analysis in Archaeology’ in Brothwell and Pollard (ed.) Handbook of
Archaeological Sciences. John Wiley pp. 685-694
Chapman, H. 2006. Landscape Archaeology and GIS. Stroud: Tempus
Clark, A. 1990. Seeing Beneath the Soil: Prospecting Methods in Archaeology. 2nd ed. London:
Batsford
Conolly, J. and Lake, M. 2006. Geographic Information Systems in Archaeology. Cambridge: Cambridge
University Press
Conyers, L. B., Lucius, J.E. 1996. ‘Velocity Analysis in Archaeological Ground-Penetrating Radar
Studies.’ Archaeological Prospection Vol. 3 pp. 25-38
Conyers, L.B., and Goodman, D. 1997. Ground-penetrating Radar: An Introduction for Archaeologists.
Walnut Creek, CA: Altamira Press
Conyers, L. B., Cameron, C. M. 1998. ‘Ground-Penetrating Radar Techniques and Three-Dimensional
Computer Mapping in the American Southwest.’ Journal of Field Archaeology Vol. 25 No. 4 pp.
417-430
Conyers, L. B., Ernenwein, E. G., Bedal, L. A. 2002. ‘Ground Penetrating Radar (GPR) Mapping as a
Method for Planning Excavation Strategies, Petra, Jordan.’ E-tiquity Number 1. ISSN 1541-5465,
http://mysite.du.edu/~lconyer/published_works.html (Accessed 21/11/08)
Conyers, L. B. 2004. Ground-Penetrating Radar for Archaeology. Walnut Creek, CA: Altamira Press
Conyers, L. B. 2006. ‘Ground-penetrating Radar for Archaeological Mapping’ in Johnson (ed.) Remote
Sensing in Archaeology: An Explicitly North American Perspective. Tuscaloosa: University of
Alabama Press pp. 329-344
66
David, A.1995. Geophysical Survey In Archaeological Field Evaluation. English Heritage Research and
Professional Services Guideline No. 1. London
David, A. 2001. ‘Overview-The Role and Practice of Archaeological Prospection’ in Brothwell and Pollard
(ed.) Handbook of Archaeological Sciences. John Wiley
De Gaetano, E. and Strutt, K. 2007. Report on the Geophysical Survey at Portus May - June 2007.
Archaeological Prospection Services of Southampton and the British School at Rome. Unpublished
report.
Fletcher, M., and Lock, R. 2005. Digging Numbers: Elementary Statistics for Archaeologists. Oxford:
Oxbow
Gaffney, C. and Gater J. 2003. Revealing the Buried Past: Geophysics for Archaeologists.
Gloucestershire, England: Tempus Publishing
Gaffney, C. 2008. ‘Detecting Trends in the Prediction Of The Buried Past: A Review of Geophysical
Techniques In Archaeology.’ Archaeometry Vol. 50 No. 2 pp. 313-336
Goodman, D., Nishimura, Y., Hiromichi, H., and Higashi, N. 2008. ‘Correcting for Topography and the
Tilt of Ground-penetrating Radar Antennae.’ Archaeological Prospection Vol.13 pp. 159-163
Goodman, D. 2008. GPR-slice User’s Manual.
Hesse, A. 1999. ‘Multi-parametric survey for archaeology: how and why, or how and why not?.’ Journal
of Applied Geophysics. Vol.41 pp. 157-168
Keay, S., Earl, G., Hay, S., Kay, S., Ogden, J., and Strutt, K. 2008. ‘The Potential of Archaeological
Geophysics: The Work of the British School at Rome in Italy.’ Conference: CNR Rome December
2008. Unpublished
Keay, S., Millett, M., Paroli, L. and Strutt, K. 2005. Portus: An Archaeological Survey of the Port of
Imperial Rome. London: The British School at Rome
Kvamme, K. L. 2003. ‘Geophysical Surveys as Landscape Archaeology.’ American Antiquity, Vol.68 No.
3 pp. 435-456
Kvamme, K. L. 2006a. ‘Integrating Multidimensional Geophysical Data.’ Archaeological Prospection, Vol.
13 pp. 57-72
67
Kvamme, K. L. 2006b. ‘Data Processing and Presentation,’ Remote Sensing in Archaeology: An
Explicitly North American Perspective, Johnson (ed.) Tuscaloosa: University of Alabama Press pp.
235-250
Kvamme, K. L. 2008. ‘Data Processing Issues in Large-Area GPR Surveys: Correcting Trace
Misalignments, Edge Discontinuities and Striping.’ Archaeological Prospection, Vol. 15 pp. 133-149
Leckebusch, J. 2003. ‘Ground-penetrating Radar: A Modern Three-dimensional Prospection Method.’
Archaeological Prospection, Vol. 10 pp. 213-240
Lillesand, T. M., Kiefer, R. W., and Chipman, J. W. 2008. Remote Sensing and Image Interpretation.
Wiley.
Neubauer, W. 2004. ‘GIS in Archaeology-the Interface between Prospection and Excavation.’
Archaeological Prospection Vol. 11 pp. 159-166
Neubauer, W. and Eder-Hinterleitner A. 1997. ‘Resistivity and magnetics of the Roman town Carnuntum,
Austria: an example of combined interpretation of prospection data.’ Archaeological Prospection
Vol. 4 pp. 179-189
Neubauer, W., Eder-Hinterleitner, A., Seren, S., and Melichar P. 2002. ‘Georadar in the Roman Civil
Town Carnuntum, Austria: An Approach for Archaeological Interpretation of GPR data.’
Archaeological Prospection Vol. 9 pp. 135-156
Piro S., Mauriello, P., and Cammarano F. 2000. ‘Quantitative integration of geophysical methods for
archaeological prospection.’ Archaeological Prospection Vol. 7 pp. 203-213
Shennan, S. 1997. Quantifying Archaeology. 2nd ed. Edinburgh: Edinburgh University Press
Van Leusen, M. 2001. ‘Archaeological Data Integration’ in Brothwell and Pollard (ed.) Handbook of
Archaeological Sciences. John Wiley pp. 575-583
Watters, M. S. 2006. ‘Geovisualization: an Example from the Catholme Ceremonial Complex.’
Archaeological Prospection Vol. 13 pp. 282-290
Wheatley, D. and Gillings, M. 2002. Spatial Technology and Archaeology: The Archaeological
Applications of GIS. London: Taylor & Francis
68
Appendix A: Maps
Map 1
69
Map 2:
70
Map 3
71
Map 4
72
Map 5
73
Map 6
74
Map 7
75
Map 8
76
Map 9
77
Map 10
78
Map 11
79
Map 12
80
Map 13
81
Map 14
82
Map 15
83
Map 16
84
Map 17
85
Map 18
86
Map 19
87
Map 20
88
Map 21
89
Map 22
90
Map 23
91
Map 24
92
Map 25
93
Map 26
94
Map 27
95
Map 28
96
Map 29
97
Map 30
98
Map 31
99
Map 32
100
Map 33
101
Map 34
102
Map 35
103
Map 36
104
Map 37
105
Appendix B: Resistance Tomography Figures
Figure 17
106
Figure 18
107
Figure 19
108
Figure 20
109
Appendix C: Text Fi lesDendrogram:
Distances between Pairs of Combined Classes
(in the sequence of merging)
Remaining Merged Between-Class
Class Class Distance
-----------------------------------------
1 2 3.076784
1 3 5.602622
-----------------------------------------
Dendrogram of z:\disser~2\discre~1\cluste~2\isocluster_sig.gsg
110
C DISTANCE
L
A
S 0 0.6225 1.2450 1.8675 2.4901 3.1126 3.7351 4.3576 4.9801 5.6026
S |-------|-------|-------|-------|-------|-------|-------|-------|-------|
2 --------------------------------------|
|--------------------------------|
1 --------------------------------------| |-
|
3 -----------------------------------------------------------------------|
|-------|-------|-------|-------|-------|-------|-------|-------|-------|
0 0.6225 1.2450 1.8675 2.4901 3.1126 3.7351 4.3576 4.9801 5.6026
111
PCA Parameters:
# Data file produced by Principal Components
#! Input raster(s):
#! ! Z:\Dissertation Final Data\Normalization Calculations\res_01
#! ! Z:\Dissertation Final Data\Normalization Calculations\mag_01
#! ! Z:\Dissertation Final Data\Normalization Calculations\gpr_01_2
#! The number of components = 3
#! Output raster(s):
#! ! Z:\Dissertation Final Data\Continuous Integration\pca3
# COVARIANCE MATRIX
# Layer 1 2 3
# --------------------------------------------------------------------------
1 4.316840e-003 4.096244e-006 2.321775e-003
2 4.096244e-006 9.987865e-004 1.571111e-004
3 2.321775e-003 1.571111e-004 2.738384e-002
#
================================================================
==========
# CORRELATION MATRIX
# Layer 1 2 3
112
# --------------------------------------------------------------------------
1 1.00000 0.00197 0.21355
2 0.00197 1.00000 0.03004
3 0.21355 0.03004 1.00000
#
================================================================
==========
# EIGENVALUES AND EIGENVECTORS
# Number of Input Layers Number of Principal Component Layers
3 3
# PC Layer 1 2 3
# --------------------------------------------------------------------------
# Eigenvalues
0.02762 0.00409 0.00100
# Eigenvectors
# Input Layer
1 0.09916 0.99507 0.00312
2 0.00589 -0.00373 0.99998
3 0.99505 -0.09914 -0.00623
#
================================================================
==========
113
