Transcript The Application of Remote Sensing and Geographic

The Application of Remote Sensing
and Geographic Information Systems (GIS) to
Prehistoric Site Location Predictive Models
Exploitation of remote sensing and GIS for real-world needs such as
land use management, urban planning and many other allied fields of
endeavor is well documented. However, there have been limited
applications of these technologies to the unique needs of properly
managing cultural resources and inadequate research in combining
remote sensing and predictive models into a cohesive tool. This
presentation is a brief overview on the development of such a tool
within the context of the NASA Goddard Space Flight Center property
in Greenbelt, MD, and the Woodland period, c. 1000 BC to 1000 AD.
Outline
The Study Area
Archaeological Field Work
What is a Predictive Model
The Building Blocks
Step by Step
Conclusion
The Study Area:
NASA Goddard Space Flight Center (GSFC)
Greenbelt, MD
Phase I Archaeological Survey
KCI Technologies, Inc (1997)
Phase II Archaeological Survey
John Milner Associates (2002)
What is a predictive model?
A predictive model is a planning and management tool. An
archaeological predictive model, as developed using modern GIS
technology and remote sensing imagery, is simply a map (or series of
maps) that shows areas of probability for the location of culturally
significant sites.
Applying New Technologies
Exploitation of geospatial information technologies, including
geographic information systems (GIS) and remotely sensed data, for
real-world needs such as land use management, urban planning and
many other allied fields of endeavor is well documented.
However, the use of these technologies is still in its infancy when it
comes to the unique needs of properly managing our collective cultural
resources.
There have been a handful of projects that have used satellite-born remote
sensors in archaeological applications, but, to date, what has been done in
this area has been limited to purely academic research and not to real-world
applications.
There are also examples, both academic and professional, of predictive
models derived through geospatial analysis techniques used for the
identification of potential prehistoric archaeological sites. However, there
has not been enough work done in combining these two areas – remote
sensing and predictive models – into a cohesive tool to meet the real-world
need of cultural resource managers.
The Building Blocks
Geomorphological and ecological factors of the
landscape.
Environmental variables of known prehistoric
archaeological site locations.
Remotely sensed imagery products.
A geographic information system (GIS) software
development environment.
Standard digital image processing techniques applied
through COTS software.
Geomorphological and ecological factors of the
Landscape
Derived from remotely sensed imagery products using standard image
processing techniques.
Will serve as an initial classification of the project area into cover-type
categories.
Primary factors:
General topography (elevation, landform, geology)
Vegetation cover (both type and density of vegetation)
Degree of modern disturbance.
Secondary factors:
Soil type
Distance to water
Slope
Aspect
Each factor, both primary and secondary, can be transformed into a
“cover-type category” that explicitly defines that one particular cover
type within the context of the research area.
Environmental variables of known prehistoric
archaeological site locations.
The locations of known prehistoric sites, as taken from earlier
archaeological research, can be used as a means of
determining which cover-type categories are most favorable for
the location of undiscovered archaeological sites within a
specific geographic region and time period.
The environmental variables of known prehistoric sites are
readily available within the individual site reports maintained by
the Maryland Historic Trust (MHT) in Crownsville, MD.
Remotely sensed imagery products.
Sensor
Panchro
matic
15m
Visible
NIR
Coverage
30m
30m
1999-2003
23.5m
23.5m
23.5m
5.8m
5.8m
5.8m
IRS 1C/1D
5m
--
--
IKONOS
1m
1m
4m
OrbView-3
1m
4m
4m
Quickbird
.67m
.67m
2.44m
Airborne DMSV
0.3m
0.3m
2m
2003present
2003present
1998present
1999present
2003present
2001present
1999present
Landsat-7 ETM+
IRS
RESOURCESAT-1
LISS-III sensor
LISS-IV sensor
Sensor
RADARSAT
Resolution
8m (Fine Beam)
Coverage
1995present
Cost
/product1
free
$2500
$2500
$2500
approx. $377
approx. $1800
$600
$20-25k
Cost
$3750
A geographic information system (GIS) software
development environment.
ESRI’s ArcGIS suite was chosen as the software development
environment as it offers many advantages.
Through its many ESRI- and third-party developed software extensions
(Spatial Analyst, 3D Analyst, and many others), ArcGIS offers the largest
range of GIS and image analysis tools at a competitive price.
ArcGIS is the COTS software currently being used by the Goddard
Environmental Team (GET) to maintain their existing geospatial data.
NASA has been standardizing on ESRI products over the past five to six
years, so its use is in line with NASA’s vision and approach to geospatial
data into the future.
Standard digital
image processing techniques applied through COTS
software.
The obvious choice for image processing software has been Leica
Geosystems’ ArcView Image Analysis Extension, which is designed for
easy integration with ESRI’s ArcGIS suite of software.
[1] State the nature of the problem.
 Define the region of interest.
 Identify the classes of interest.
[2] Acquire remote sensing and ground reference data.
 Select remotely sensed data based on defined criteria.
 Obtain initial ground reference data based on a priori knowledge of the
study area.
[3] Process remote sensor data to extract thematic information.
 Radiometric correction and geometric rectification.
 Select appropriate image classification logic and algorithm (supervised,
unsupervised, hybrid).
 Extract data from initial training sites using most bands.
 Select the most appropriate bands using feature selection criteria (graphical
and/or statistical).
 Extract training statistics from final band selection.
 Extract thematic information.
[4] Error evaluation of the classification map.
1. Obtain additional reference test data based on a posteriori knowledge of the
study area and a stratified random sample.
2. Assess statistical accuracy of the classification map.
Putting it All Together
COTS Software Development Package
Step By Step
Develop basic geomorphological and environmental
cover-type maps from remotely sensed data.
Catalog the environmental variables of known
prehistoric sites.
Develop preliminary probability maps.
Review initial results and adjust model if necessary.
Develop basic cover-type maps from remotely sensed
data
[1] Acquire the Data
A “cover-type map” is vector data created within GIS software
that shows the distribution of a single geomorphological or
environmental factor of prehistoric site location.
The primary multispectral data source should meet the following
criteria:
High ground resolution so that environmental variables of
prehistoric site location can be properly identified.
Availability of panchromatic, visible light and near-infrared (NIR)
imagery products.
Proper coverage of the proposed project area both geographically
and temporally.
Reasonable cost.
Sensor
Panchro
matic
15m
Visible
NIR
Coverage
30m
30m
1999-2003
23.5m
23.5m
23.5m
5.8m
5.8m
5.8m
IRS 1C/1D
5m
--
--
IKONOS
1m
1m
4m
OrbView-3
1m
4m
4m
Quickbird
.67m
.67m
2.44m
Airborne DMSV
0.3m
0.3m
2m
2003present
2003present
1998present
1999present
2003present
2001present
1999present
Landsat-7 ETM+
IRS
RESOURCESAT-1
LISS-III sensor
LISS-IV sensor
Sensor
RADARSAT
Resolution
8m (Fine Beam)
Coverage
1995present
Cost
/product1
free
$2500
$2500
$2500
approx. $377
approx. $1800
$600
$20-25k
Cost
$3750
Develop basic cover-type maps from remotely sensed
data.
[2] Process the Data
The Quickbird imagery products would then be brought into the image
processing software environment (Leica’s Imagery Analyst Extension)
to transform the remotely sensed data into thematic information
regarding the geomorphological and ecological cover-types in the
project area.
[1] State the nature of the problem.
 Define the region of interest.
 Identify the classes of interest.
[2] Acquire remote sensing and ground reference data.
 Select remotely sensed data based on defined criteria.
 Obtain initial ground reference data based on a priori knowledge of the
study area.
[3] Process remote sensor data to extract thematic information.
 Radiometric correction and geometric rectification.
 Select appropriate image classification logic and algorithm (supervised,
unsupervised, hybrid).
 Extract data from initial training sites using most bands.
 Select the most appropriate bands using feature selection criteria (graphical
and/or statistical).
 Extract training statistics from final band selection.
 Extract thematic information.
[4] Error evaluation of the classification map.
1. Obtain additional reference test data based on a posteriori knowledge of the
study area and a stratified random sample.
2. Assess statistical accuracy of the classification map.
Catalog the environmental variables of known
prehistoric sites.
Acquire Data
Detailed archaeological site data for the local region can be acquired from
the Maryland Historic Trust (MHT).
MHT is the same source of archaeological site data used in similar
predictive model projects, such as a predictive model developed for the
Aberdeen Proving Grounds (APG) in 1996 (Wescott and Brandon, 2000).
Process Data
Categorize the known prehistoric site locations with respect to their sitetypes (i.e., shell midden, lithic scatter, and others) and environmental
attributes associated with individual site location (such as soil type, distance
to water, wetland vs. dry land, slope, aspect, etc.).
The result of this step will be tabular data (Excel tables) relating site-type to
environmental attribute, providing a basis for the extrapolation of potential
archaeological site locations from the cover-type maps derived from the
remotely sensed data.
Develop preliminary probability maps.
[1] Distribution and frequency of
known prehistoric sites
Distribution and Frequency Maps
Associate tabular site-type data (derived from the individual archaeological
site reports) with the vector cover-type data (derived from the remotely
sensed imagery) within the GIS software environment (ArcGIS).
The process for creating such thematic maps from existing tabular and
vector data is a standard function of the GIS toolkit and can be found in any
textbook (Demers, 2000) or software user’s manual.
Importance
It is an accepted approach, found in many recent archaeological projects
using predictive models (Wescott and Brandon, 2000), that what is true for
the larger region can be assumed to be true for a representative subset of
that region as far as settlement patterns go.
As an example of this reasoning, it could be said that if the frequency of
prehistoric lithic scatter sites within moderately dense, undisturbed
woodland throughout the Maryland Coastal Plain Province (of which GSFC
is a subset) is roughly 1 per square mile (a purely hypothetical number),
then one would expect somewhere around 19 prehistoric lithic scatters
within the 19 square miles within the GSFC property line that is also
moderately dense, undisturbed woodland.
Develop preliminary probability maps.
[2] Probability of Unknown
Prehistoric Sites
The environmental parameters of known prehistoric site locations along
with the level of modern disturbance will serve to derive the first cut
probability maps by assigning weighted values to each cover-type.
The result of this weighting will be a series of maps that visually display
the probability of locating a specific prehistoric site-type within a covertype category.
Assumption
Areas associated with the cover-type “modern/disturbed” have a very low
likelihood of containing archaeological remains of cultural significance
because of their dubious cultural origin.
Areas that are not “modern/disturbed” have a greater likelihood of containing
undisturbed prehistoric sites, which should be reflected in the resulting
probability maps.
Note: The weight value range shown here is a somewhat arbitrary range and
won’t be clearly identified until the task of cataloging the archaeological data
has been completed. From other sources (Wheatley and Gillings, 2002;
Wescott and Brandon, 2000; Jensen, 1998), this range is assumed to be a close
approximation of what will be determined during the course of this project.
Certain environmental parameters (such as elevation, distance to water, soil
type, etc.) of site location should give modifiers to the above weighted values,
shifting a cover-type up or down on the scale.
Certain environmental factors associated with preservation processes (such as
wetland vs dry land) should give modifiers to the above weighted values,
shifting a cover-type up or down on the scale.
Value
0
1-5
-2 - +2
-2 - +2
Develop preliminary probability maps.
[3] Visibility Maps
Visibility maps can show how easy or hard it is to identify unknown
prehistoric sites both by remote sensing survey and archaeological field
survey. Visibility is determined from the primary factors of vegetation
density, land cover type, topography, elevation, geology, and landform,
and so can be derived solely from remote sensing imagery.
Visibility
Factor
Vegetation
Density
Land Cover
Type
Elevation
Geology
Landform
Affect
Dense vegetation cover can preclude certain remote sensing survey
methods as well as hide archaeological remains from ground surveys.
Conversely, sparse vegetation cover would be beneficial for both
types of survey.
Certain land cover types can make survey more difficult. Wetlands
generally promote good preservation of the archaeological record,
but are difficult to survey both physically on the ground as well as
from remote sensing options. Deserts are also beneficial to
preservation of remains – ground survey can be hard in this terrain,
but the nature of deserts make remote sensing methods very
appealing.
Extremes in elevation can make ground survey more difficult or
impossible, though it does not have as much effect on remote sensing
surveys.
Though remote sensing devices can easily discern exposed rock
formations, it is not useful in determining if these formations where
used by past peoples (i.e., you can see the rock, but not the paintings
on the rock).
The shape of the land itself may prevent ground survey even if the
probability of finding an archaeological site is high.
Review initial results and
adjust model if necessary.
It is necessary to provide statistical validity to the resulting probability
maps by comparing them to the known locations of prehistoric sites
within the GSFC property line.
There are two archaeological field surveys at GSFC that will provide the
necessary information to perform this check – a 1997 Phase I survey (KCI
Technologies, 1998) and a 2003 Phase II survey (John Milner Associates,
2004).
This will also help further refine the probability maps in that anything
about the existing site locations that does not mesh well with our model
will serve as an indication of something in the model that needs to be
adjusted. Earlier stages can be revisited to adjust the probability model
and resulting maps.
Conclusion
The process outlined in this presentation is easily performed within the
scope of the technology available today. Many other disciplines have
been successfully using this general approach to improve their
management of resources through the use of remotely sensed imagery
and GIS techniques – and there is no reason why cultural resource
managers can not benefit in the same way.
The ground resolution of the remotely sensed data has traditionally
been the limiting factor with using remote sensing imagery for
archaeological purposes. This is no longer absolutely true with the new
breed of high resolution platforms currently offering imagery products
commercially, with ground resolutions quickly approaching the submeter level.
The day will come when the techniques shown in this presentation are
part of the standard toolkit of cultural resource managers and field
archaeologists – it is simply a matter of how and when these tools are
adopted.
Bill Dickinson Jr.
Dickinson Jr., William B., “A Remotely-Sensed Decision-Support Tool
For Facilities Planning” (NASA SBIR proposal), 2005.
Dickinson Jr., William B., “Proposed use of Vegetation Indices at
Goddard Space Flight Center” (NASA internal document,
NNG04AZ01C), 2004.
Dickinson Jr., William B., “Cultural Resource Management: An
Application of Remotely Sensed Data and Advanced Image Processing
Technologies” (NASA SBIR proposal), 2004.
Dickinson Jr., William B., “Archaeological Predictive Models: A Look
Into Commercial Potential” (white paper), 2003.
Dickinson Jr., William B., “Safety and Environmental Branch GIS
Planning Document” (NASA internal document, NAS5-99001), 2001.
Bibliographic References
JMA, Inc., “Phase II Archaeological Field Survey of Goddard Space Flight Center.” 2003.
Wheatley, David and Gillings, Mark, “Spatial technology and archaeology: The archaeological
applications of GIS.” 2002.
Demers, Michael N., “Fundamentals of Geographic Information Systems.” 2000.
Gillings, Mark editor, “Geographical information systems and landscape archaeology.” 1999.
Spikins, Penny, “GIS Models of Past Vegetation: An Example from Northern England, 10,000-5000 BP.”
1999.
Kickert, R.N., Tonella, G., Simonov, A., and Krupa, S.V., “Predictive modeling of effects under global
change.” 1999.
Renfrew, Colin and Bahn, Paul, “Archaeology Theories, Methods and Practice.” 1998.
KCI Technologies, Inc., “Phase I Archaeological Field Survey of Goddard Space Flight Center (GSFC).”
1997.
Rivett, Paul, “Conceptual data modeling in an archaeological GIS.” 1997.
Schmidt Jr., Martin F., “Maryland’s Geology.” 1997.
Conyers, Lawrence B. and Goodman, Dean, “Ground-Penetrating Radar: An Introduction for
Archaeologists.” 1997.
Jensen, John R., “Introductory Digital Image Processing: A Remote Sensing Perspective.” 1996.
Lyons, Thomas R. and Mathien, Frances Joan editors, “Cultural Resources Remote Sensing.” 1980.
Aikens, C. Melvin et al, “Remote Sensing: A Handbook for Archaeologists and Cultural Resource
Managers, Basic Manual Supplement: Oregon.” 1980.
Some other Research on
Archaeological Predictive Models
Madry, Scott, “GIS and Remote Sensing for Archaeology: Burgundy, France.” 2004.
Craig, Nathan and Aldenderfer, Mark, “Preliminary Stages in the Development of a RealTime Digital Data Recording System for Archaeological Excavation Using ArcView GIS 3.1.”
ESRI Journal of GIS in Archaeology, Volume 1, April 2003.
Johnson, Ian and Wilson, Andres, “The TimeMap Project: Developing Time-Based GIS
Display for Cultural Data.” ESRI Journal of GIS in Archaeology, Volume 1, April 2003.
Comer, Douglas C., “Environmental History at an Early Prehistoric Village: An Application
of Cultural Site Analysis at Beidha, in Southern Jordan.” ESRI Journal of GIS in
Archaeology, Volume 1, April 2003.
Clement, Christopher O., De, Sahadeb, and Wilson Kloot, Robin, “Using GIS to Model
and Predict Likely Archaeological Sites.” 2002.
Sherbinin, Alex de et al., “A CIESIN Thematic Guide to Social Science Applications of
Remote Sensing.” 2002.
Burson, Elizabeth, “Geospatial Data Content, Analysis, and Procedural Standards for
Cultural Resources Site Monitoring.” U.S. Army Corps of Engineers, 2001.
Wescott, Konnie L. and Brandon, R. Joe, “Practical Applications of GIS for
Archaeologists: A Predictive Modeling Kit.” 2000.