See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283863279

A geographic information systems approach to understanding potential

prehistoric and historic travel corridors

Article  in  Southeastern Archaeology · June 2003

CITATIONS READS

37 257

2 authors, including:

Thomas Whitley
Sonoma State University
26 PUBLICATIONS   239 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Georgia View project

All content following this page was uploaded by Thomas Whitley on 08 December 2017.

The user has requested enhancement of the downloaded file.

A GEOGRAPHIC INFORMATION SYSTEMS APPROACH TO UNDERSTANDING
POTENTIAL PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Thomas G. Whitley and Lacey M. Hicks to cross large expanses of terrain or to move goods from
region to region) as well as secondary pathways (those
used to reach local resources).

As part of a Geographic Information Systems (GIS) predictive

model of the proposed Northern Arc highway corridor north Theoretical Perspective
of Atlanta, plausible associations were hypothesized between
key environmental factors and archaeological localities. One GIS analysis has evolved as a tool for archaeologists
byproduct of the investigation was the creation of a GIS and anthropologists principally over the last decade
surface representing ‘‘travel friction’’ and a model of potential and a half (Aldenderfer and Maschner 1996; Allen et al.
travel arteries. By roughly approximating travel costs, the 1990; Gillings et al. 1999; Lock 2000; Lock and Stancic
model identifies the lowest cost travel vectors in the region. 1995; Wescott and Brandon 2000). Substantial applica-
These are illustrated and compared with known archaeolog- tion of GIS analysis has also occurred in other dis-
ical sites from different temporal periods. ciplines, such as theoretical and applied geography,
complexity theory, behavioral psychology, artificial in-
telligence, and computer science. For instance, the
During initial background research for the proposed dynamic cognitive mapping of spatial areas and sub-
Northern Arc (a highway corridor in the Piedmont sequent temporal applications of GIS have generated
north of the city of Atlanta, Georgia), Brockington and much discussion in geography (e.g., Downs and Stea
Associates archaeologists compiled a series of digital 1977; Hirtle and MacEachren 1998; Langren 1992),
datasets for use in creating a Geographic Information which is influencing the debate in archaeology (e.g.,
Systems (GIS) -based archaeological predictive model. Gaffney et al. 1996; Maschner 1996; Wheatley 1993;
Although the research presented here makes use of some Whitley 2002; Witcher 1999; Zubrow 1994). Similarly,
of these data, it does not detail any of the methods, neural network and fuzzy modeling applications are
analyses, or results of the Northern Arc project. Rather, just now seeing fruition in archaeology (e.g., Barcelo
our goal is to focus on the potential uses of such data- and Faura 1997; Bell 1998; Bell and Jantz 2002; Crescioli
sets for exploring, identifying, and interpreting more et al. 2000; Niccolucci et al. 2001; Reeler 1999).
complex sociocultural themes using GIS software. Processual North American archaeology, though, has
This is foremost an exploratory examination of the been less welcoming of interdisciplinary spatial inter-
data; our primary goal is to generate questions rather pretations than has European archaeology with its
than answer them. With such a study area as this— emphasis on landscape analysis. This is largely a factor
which, like many in the Southeast, has had very differ- of the narrow definition of scientific explanation still
ent levels of archaeological survey over the years—we accepted within the processual school (Salmon 1998:
need not limit our explanations of human behavior by 347–363). Without getting into the extensive debate
relying on incomplete and biased datasets. So our between the positivist (processual) and contextual (post-
secondary goal is to encourage those who routinely deal processual) schools (cf. Binford 1989; Hodder 1986),
with prehistoric Piedmont occupations to consider how suffice it to note that both perspectives have some
our questions, and others like them, may be addressed serious limitations. These limitations are only (or, at
with GIS analyses. least, best) overcome with the acceptance of a probabi-
We first consider the theoretical nature of our research listic interpretation of causality (Salmon 1984, 1998),
in light of advances made in GIS applications to a statistical relevance model of scientific explanation
archaeology. Second, we discuss the original datasets, (Salmon 1971), and a logical mechanistic approach to
how they were compiled and standardized, and how understanding cognitive decision making (cf. Renfrew
additional data were derived. This includes a cursory 1994).
examination of the basis and techniques of cost The approach advocated here accepts the potential
modeling and creating friction surfaces in archaeology. for complex (dynamical) interactions, envisions terrain
We also present the results of several complex GIS models and environmental variables as fuzzy parame-
analyses that indicate how and where prehistoric ters acting within cumulative and individual cognitive
populations may have traveled in the region. Ultimately, landscapes, and suggests methods in which logical
this study focuses on primary travel arteries (those used arguments about cognitive choices are derived from

76
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Figure 1. Project area location.

real-world empirical correlations. As such, it represents cognitively and behaviorally used the same variables.
an implementation of Renfrew’s (1994) cognitive-proc- We in no way suggest that these few simple variables
essualism. Though Renfrew rejects the environmental are the only ones worth scrutinizing. There are perhaps
deterministic stance of much empiric correlative analy- a thousand different ways in which one might imagine
sis, he also singles out antiprocessualist rejection of the a spatial relationship between elements of cognition
scientific method as distinctly ‘‘confused, and ultimately and decision making. Our effort here is to stimulate
unhelpful’’ (Renfrew 1994:4). An examination of the discussion on such ideas within the greater Piedmont
nature of cognition, not contextual interpretations, research realm. Perhaps additional variables may
should be the focus of a cognitive-processual approach. eventually be digitized and combined into representa-
Understanding the ancient mind is to come from tions of prehistoric cognitive elements, or at least some
mechanistic explanation, not from environmentally proxies may be envisioned and employed.
deterministic correlative explorations or from specula-
tive postmodernist attempts to infer ‘‘meaning’’ from
the archaeological record.
The use of strictly environmental parameters within The Datasets
this interpretation is not an adoption of, or relegation to,
environmental determinism. Rather, the datasets gener- The study area is a 90-
-25-km (56-
-16 mi) rectangle
ated by our predictive modeling project were, in point in North Georgia (Figure 1) including the entirety of 16
of fact, largely ecological in nature. Our attempt to USGS 7.5-minute quadrangle maps covering almost all
transform those data into representations of travel effort of Bartow, Cherokee, and Forsyth Counties. All analysis
and trajectory preferences begins the process of using was carried out with ArcViewÓ 3.2 (ESRI 2000). The
our extant environment to logically deduce and un- area is a mix of typical southern Piedmont and North
derstand ways in which prehistoric people may have Georgia mountains habitats (Hodler and Schretter

77
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

1986). Portions of the Chattahoochee and Etowah Several derived datasets were created from the
drainages lie within the study area. original DEMs, including a hillshade model (showing
The primary base map used for the majority of our ambient directional morning sunlight), a slope surface
analysis consists of the combined 30-m digital elevation (showing percent of slope as a function of Z-values),
models (DEMs) corresponding to the 16 USGS 7.5-minute and a hydrological model (showing streams as a function
quadrangle maps. The DEMs consist of grid files (raster of projected flow direction and accumulation, assuming
data) with elevations (Z-values) spaced evenly across the even precipitation across the project area). Figure 2
surface every 30 m (approximately 100 ft). These data illustrates the DEMs overlaid with the hydrological
were originally interpreted from stereo aerial photo pairs datasets (polygons and polylines). Some of the derived
keyed to the 7.5-minute quadrangles, corrected, and geo- data (such as slope surface, plus flow direction and flow
referenced by the USGS. accumulation models) were then used as original data-
Additionally, hydrological data were compiled for sets for the extraction of friction surfaces.
the study area, including lake polygons (principally, the The dataset of previously recorded archaeological
two large reservoirs in the study area, Lakes Allatoona sites was compiled from information in the State of
and Lanier) and stream polylines. The reservoirs, in fact, Georgia Archaeological Site Files (maintained by the
were a source of elevation error to be considered while University of Georgia in Athens). The dataset (complete
undertaking the analyses. The DEMs show surface through July 2001) consists of 1,966 archaeological sites
topography as it existed when the aerial photographic previously recorded for the region from all temporal
data were compiled, meaning that the water surfaces periods. Some field verifications have been carried out
of both Lake Allatoona and Lake Lanier were used to to confirm site location data, and sites with question-
represent the elevation. We know, however, that prior to able locations were removed from the analysis. For
lake creation, the actual surface elevations of the ground analysis, only sites with clearly identifiable temporal
were lower (lying beneath the water today). Taking this associations (Paleoindian, Archaic, Woodland, Missis-
into account, projected pathways through either of these sippian, Historic Indian, and Historic European) and
reservoirs are much less accurate than those where no site types (e.g., camp, village, artifact scatter) were used.
artificial surfaces occur. This, in combination with removal of the Lake Alla-
This error could have been addressed by going to toona and Lake Lanier sites, resulted in a total database
the original topographic maps for the region, extract- of 908 temporally identifiable sites usable for analysis
ing those areas within the current lake polygons, and (Figure 3).
transferring the original elevation values to the modern
dataset (sampled at the same resolution). Conversely,
the modern lake polygons could have been used to clip Analysis and Interpretation
the grid dataset, and with the addition of a limited
number of original elevation values, an interpolation When presented with the datasets compiled for the
surface could have been created. Since neither of these Northern Arc project, our initial considerations were
areas was an issue for the proposed Northern Arc to think about the region as an inherently exploitable
project, there was no funding available to rectify the resource and habitation area, as well as a regional
lake surface elevations during the predictive modeling crossroads for the movement of people and goods.
process. For this ancillary project, it was decided that Aside from part of the Etowah site complex located in
interpolated elevation values in reservoir areas would the southwestern portion of the study area, there are no
not provide the same resolution as the rest of the digital strongly clustered sites that are not attributable to
data, and accurately transferring original elevation specific survey tracts. There are, however, five major
values from pre-reservoir maps would have been trails that crossed the region during the Historic period:
prohibitively expensive. Hightower Trail (‘‘Hightower’’ is probably a phonetic
In addition, sites identified during archaeological corruption of ‘‘Etowah’’), New Echota Trail, War-
surveys carried out for both reservoirs prior to their woman’s Trail, Alabama Road/Tugaloo Trail, and the
creation had to be removed from the analyses, even Cherokee Federal Road. Information on the identities
though they present a wealth of material for cultural and locations of these known pathways can be found in
interpretations. These intensive surveys overemphasize numerous maps provided by John Swanton (1922),
two small portions of the entire study area, and projected particularly those by Mitchell (1755) and Purcell (1770).
travel arteries there may be inherently less accurate than Hargrett Rare Map Collection Map #1779:N42 (1779) is
neighboring ones. Since this exploratory project focuses also an excellent early source map for trail locations, and
on large-scale interpretation of projected pathways, the Hemperley’s (1978) compilation was the primary source
reservoir areas (with their faulty elevation datasets and used by Hodler and Schretter (1986:68) for their overlay
recorded sites) were deleted from the analysis. This was of known trails on a rectified, scaled, relief map of
the easiest and least expensive option. Georgia.

78
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Figure 2. Digital elevation model (DEM) with hydrology layers.

Our first tasks were to evaluate the nature and the cost of reaching any point or resource is a function of
patterns of travel cost, recreate the most efficient travel the effort required to traverse the distance in between.
arteries through the region, and see how they compare This, in effect, is known as cost distance. A GIS analysis
with both known site distributions and known Historic can evaluate cumulative travel effort by multiplying the
period trails. We hoped this would provide an under- linear distance separating two points by the friction
standing of the time depth of historic pathways while values for each land unit or pixel between them. The
also identifying other potential travel arteries. In order key to an accurate cost distance evaluation is choosing
to accomplish this, we ran a least cost path analysis. appropriate friction values for every pixel. In this case
Although ideas such as differential terrain costs, travel every 30-m land unit—or devising a formula to calculate
effort, and conditional spatial benefits were conceptual- automatically the friction values as a function of some
ized long ago by Industrial Location (Weber 1929) and existing dataset(s).
Central Place (Christaller 1935) theorists, their applica- The simplest form of a friction surface (or an ac-
tion to real terrain models was initiated by Warntz cumulation cost surface, cf. Warntz 1957) is a slope model.
(1957, 1965) and Lindgren (1967). GIS-based least cost ArcViewÓ 3.2 includes a built-in procedure to extract
path analysis has become commonplace in analytical the slope from a DEM, as long as the map units and
geography and has had several applications in archae- the Z-values are in the same format, and the resulting
ology (e.g., Anderson and Gillam 1997, 2000; Bellavia output can be derived in either degree of slope or per-
2002; Hartley and Vawser 1998; Jensen 2002; Whitley cent. Theoretically, the cost of traversing a land unit will
2000, 2002). increase as the slope increases. In the past, many studies
have used slope as a direct reflection of travel cost.
Although that approach may not be inherently wrong,
Primary Pathways
the range of slope in the region may affect the results,
The initial step in least cost path analysis is to create since a highly dissected environment may overempha-
a friction surface. In a GIS environment the distance to size the cost of traversing a short slope in a direct line in
any given point can be evaluated as a straight linear favor of a more circuitous route where no slope is crossed.
measurement. Typically, many archaeological GIS This can be moderated by transforming the slope
studies have used direct linear distance to water as model with a hiker function (e.g., Tobler 1993; Hartley
a mitigating factor in site location or predictive models and Vawser 1998), a mathematical formula that uses
(e.g., Warren 1990; Wescott and Kuiper 2000). In the real the slope value to calculate a walking speed. Walking
world, however, we know that site locations are not speed may more directly equate with travel cost since
limited by a direct linear distance to anything. Rather, it moderates extreme values and can be translated into

79
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

Figure 3. Digital elevation model (DEM) with known archaeological sites.

caloric expenditure. The slope surface created from the one in which the highest calculated walking speed was
original DEMs was transformed by application of the given the value of 0 and the lowest calculated walking
hiker formula into walking speed (assuming a 1-m speed the value of 1; everything in between scored a
stride across the 30-m land units): proportional decimal value (to two places).
This had two advantages. First, the standardized
Walking Speed (kmph) friction surface could then be weighted differentially for
¼ 6 expð3:5  jðSlope percent þ 0:05ÞjÞ application in predictive formulas (as was the original
intent of the surface). Second, it allowed us to disregard
Of course, we also know that travel cost reflects not the nature of regional caloric intake, types of food
only slope but also vegetative cover, presence of wet- carried on cross-country travel between settlements,
lands, travel method, group size, and climatic conditions, availability of resources, and so on. We assume that
among others factors. Tobler (1993) uses a simple regional resource utilization is reflected in the spatial
mathematical transformation to distinguish between on- analysis in a general or relative way, and that no
path, off-path, and horseback travel (multiplying the particular trail was more amenable to a different travel
calculated walking speed by factors of 1.0, 0.6, and 1.25, speed (engendered by group size, travel conditions, or
respectively). Since we cannot recreate most of these available foodstuffs, etc.) than any other. It is not our
factors without much more extensive research, no intention to distinguish between travel arteries on the
transformative values were established. Instead, we as- basis of caloric potential, though we acknowledge that it
sumed that all travel, although reflected by calculated may have greatly affected travel potential.
potential walking speed (based on slope), occurred along With the limited data on hand, we added only one
established pathways, not off-path or on horseback. additional variable to the final friction surface: flow ac-
This raises the issue of caloric expenditure as well. cumulation, which is a surface derived from hydrological
Since we did not wish to convey the idea that the least modeling. The built-in ability of ArcViewÓ to use a
cost paths were dependent on any particular group size DEM to develop a model of drainage flow is inherently
or travel method but were, in fact, generic relative least useful for evaluating travel costs. In theory, it requires
cost evaluations, we chose not to transform the hiker a greater expenditure of effort to cross a large river
function into an expression of caloric distance. Instead, than a small ephemeral stream. The flow accumulation
they were transformed into generic cost units, with the function takes each land unit, evaluates adjacent land
highest walking speeds reflecting the lowest relative units, determines in which direction water would flow
costs and vice versa. In other words, the walking speed (based on the slope and the Z-values), and calculates
values were transformed from a km-per-hour value to the total number of other land units that drain into

80
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Figure 4. Final friction surface.

each one. By administering a threshold value, a model path analyses for each of the relevant origin/destination
of streams is created. combinations. A cost distance evaluation was carried
The flow accumulation surface itself provides a useful out for each origin point dataset; which was in turn used
indicator of travel effort if those costs are translated into to calculate least cost paths to each destination point
values comparable to the cost units used for walking dataset. These calculations were carried out for paths
speed. This was done for the model and added to the running north to south, south to north, east to west, and
hiker function, resulting in the ultimate friction surface west to east. All least cost paths were then combined
(Figure 4). Similarly, it would be possible to transform into a single set of polylines reflecting the most efficient
the flow accumulation surface into one of water travel trajectories for traveling through the region from any
costs, given that we have the ability to model flow levels single point on the perimeter to any other point on the
and direction. This, however, was not done for this perimeter of the study area (Figure 5). These we labeled
study, and all prehistoric travel is assumed to be on foot. primary pathways.
The final friction surface (as presented in Figure 4) Due to the edge effect, not all of the terminal ends of
illustrates the total cost of traversing each land unit in the primary pathways are presented in Figure 5. In the
the study area. This cost surface forms the mathematical mathematical routine performed by ArcViewÓ, as
core of the least cost path analysis; with it we can mea- the calculations approach the edge of the target area the
sure how much relative effort is expended traveling analysis will sometimes create artificial lengths of
between any two points in the environment. In the GIS pathway that follow the edge of the project area,
this involves creating an origin and a destination point because slope direction is either assumed to be flat at
to be evaluated. Because we are concerned initially the very edge or can only be calculated from a smaller
with travel arteries through the region, we established quantity of pixels (since the exterior is unknown).
a baseline of origin and destination points around the Several of the 1-km origin/destination points were
perimeter of the study area spaced at 1-km intervals. connected by pathways around the exterior of the study
Additional origin/destination points were created area rather than by short lengths across the actual
spaced at 2 km, 5 km, 10 km, 20 km, and 50 km. Since terrain. These edge-effected paths were removed for
a least cost path analysis is influenced by the specific clarity and did not influence the vast majority of the
locations of the origin and destination, it was felt that study area.
different origin/destination point datasets would allow In three locations, additional origin/destination
greater detail and comprehension. points were also created to identify possible primary
The ‘‘Cost Distance Tools’’ extension for ArcViewÓ pathway shortcuts, alternative branches of major trails.
3.2 was used to run both cost distance and least cost These alternative origin/destination points were placed

81
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

Figure 5. Primary pathways (least cost paths).

in areas that suggested that there were very slight a heavily utilized travel artery. Understanding whether
thresholds between long stretches of uninterrupted least ‘‘shortcuts’’ or alternate travel arteries were well known
cost paths. For example, along the length of the pathway gives a great deal of insight into regional resource
that we later designated as the probable location of the familiarity, cultural conditioning, and ultimately the ex-
Tugaloo Trail, a long stretch lies adjacent to and west of pression of prehistoric cognitive landscapes.
a major ridgeline (actually a series of connected ridges)
running southwest to northeast. There is a similarly
Secondary Pathways
long valley occupying the east side of the ridgeline.
Intuitively we recognized that it could, in fact, form an In order to understand the regional distribution of
alternate (shorter) trajectory to reach the same destina- secondary trails—those that show a relationship not
tions. The GIS least cost path algorithm, though, was to travel through an area but to utilization of local
not refined enough to traverse a very small intervening resources—additional analyses were carried out. Sec-
high slope area. To draw out this possible alternate path- ondary pathways are considered here to be those short
way, we placed an origin/destination point along the travel arteries that allow access to a local resource area.
GIS calculated pathway at the north and south ends of Though clearly the utility of any resource area will be
this long valley and ran the least cost path analysis a function of a series of environmental and cultural
again. The result was an additional pathway that fol- constraints (see Whitley 2000), no specific resource
lowed the valley on the east side of the ridge. attractors (such as lithic sources) were modeled for this
This potential to miss possible alternate trails suggests study. Since the goal of research, at this point, is ex-
to us that you cannot always rely solely on a GIS or ploratory, resource uniformity was assumed across the
statistically generated interpretation. Second, it suggests region. Thus, we were interested in observing potential
that reliance might have been not only on the least local pathways without regard to which resources were
cost pathway but also often on low (not least) cost being used. This approach generates a series of pathways
alternatives. Clearly, humans intuitively recognize the that reflect travel effort as related to the primary arterials
value of alternate pathways and keep in mind long and makes no presumptions about what resources were
term, complex, cumulative cognitive landscapes reflect- preferred, only that there must be resources available
ing trajectory choices. The best possible use of GIS as within the general area. Secondary pathways, therefore,
a tool is to recognize the inherent limitations that will are the least cost trajectories by which to access resources.
always require a little bit of interpretation to under- A built-in routine of ArcViewÓ makes identifying
stand. Future researchers could investigate sites along secondary pathways particularly easy. Once primary
this alternate pathway to determine if, in fact, it was pathways are established (as described above), a cost

82
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Figure 6. Cost topography (cost distance to primary pathways).

distance evaluation is carried out for the project area When overlaid on the simulated elevation of the cost
using the same friction surface and primary pathways topography, the secondary paths fit very well into the
as the origin. The resulting cost distance surface is then artificial valleys and hollows, with the costsheds form-
transformed into a cost topography (Figure 6), a con- ing along the ridgetops. When, overlaid on the actual
ceptual model that uses cost distance as a simulator topography, though, it is easy to see that primary paths
of elevation. The primary pathways form the origin typically lie within and along major waterways, whereas
points and would therefore represent ‘‘sea level’’ in the secondary paths lie along small tributaries and ridges
cost topography. The remainder of the study area has running perpendicular to the primary paths. Costsheds
‘‘elevation’’ (Z-values) calculated as a function of its cost do not necessarily conform to what we might see as
distance to the nearest primary pathway. The highest ‘‘natural’’ boundaries, such as watershed boundaries.
points in the surface indicate the locations to which
travel cost is highest. Once the cost topography has been
Temporal Depth
established, a hydrological analysis fairly easily extracts
the simulated ‘‘streams.’’ In this case, streams represent How do the primary pathways compare with known
not the flow of water but the least cost paths into the historic trails and the distributions of archaeological
local uplands from the primary pathways—in other sites? The general locations of historically known trails
words, the secondary pathways (Figure 7). An example acquired from historic maps (as compiled in Hodler
of pathway extraction from a cost topography was and Schretter 1986) were digitized to fit the study area
presented by Douglas (1994) and used specifically in an (Figure 8). More precise estimates of their locations were
archaeological adaptation by Bellavia (2002). extracted by buffering the digitized lines within the
Unlike primary pathways, secondary paths do not primary pathways dataset. Figure 8 shows the probable
travel from north to south or east to west, but branch extent of the Tugaloo, Hightower, New Echota, and
outward from the primary paths, much like dendritic Warwoman’s Trails and the Cherokee Federal Road
streams. Similar to dendritic streams, each set of as a function of least cost expenditure (the primary
secondary paths also forms a ‘‘watershed’’ or costshed, pathways). Like most pathways, we are assuming that
a conceptual interpretation of a boundary between multiple arteries represent what was historically iden-
potential resource areas in which the cost of travel tified as a single trail.
exceeds a certain threshold (see Figure 7). Though the An association between sites of known temporal
threshold is, at this point, an arbitrary value, some periods and each of these pathways would suggest a time
specific cost thresholds may positively correlate with depth beyond our direct knowledge from the Historic
cultural boundaries if they were to be examined in detail. period (Table 1). None of the relationships observed in

83
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

Figure 7. Secondary pathways and costsheds (pseudo-hydrological model).

Table 1 are statistically significant (with the exception of indian and Archaic populations frequently using the
the associations between the Tugaloo Trail and sites from travel artery), whereas the Hightower and New Echota
each time period other than the Paleoindian) due to the Trails may not have become well used until at least the
small dataset of known archaeological sites, although Woodland period.
one trend is suggestive. There do appear to be several additional trails asso-
Half of the 10 (useable) Paleoindian sites in the region ciated with both Paleoindian and Archaic sites (as well
are within 500 m of some portion of the Tugaloo Trail, as later occupations) that run north-south and are not
and one other is within 500 m of the Cherokee Federal identified Historic period trails (Figure 9). These primary
Road. The remaining Paleoindian sites are not asso- pathways may have been the most commonly used
ciated with Historic trails. We are not suggesting that north-south arteries prior to the development of Etowah,
the association with Paleoindian sites is particularly the Hightower or New Echota Trails, and the Cherokee
impressive, since the sample is so small, but the trend Federal Road. More investigation (including historic
does suggest that a concerted effort to identify documentation) needs to be done to evaluate the role of
Paleoindian sites and evaluate their relationship to these other primary pathways.
natural pathways could illuminate the issue.
A stronger association exists between the Tugaloo
Temporal Associations with Primary Pathways
Trail and known Archaic sites (24 percent within 500 m,
37 percent within 1 km). Many known Woodland and On a more general level, we evaluated relationships
Mississippian sites likewise occur within 500 m (32 between known sites and all of the primary pathways
and 39 percent, respectively) or within 1 km (42 and 61 to identify patterns in overall settlement. First, the cost
percent, respectively) of the Tugaloo Trail. In contrast topography was reclassified to a series of 100 cost unit
to the Paleoindian and Archaic sites, Woodland and (cu) categories (i.e., 0–100 cu to 900þ cu). For the
Mississippian sites most frequently occur near the reasons listed above, we decided that caloric expendi-
intersection of the New Echota Trail and the Tugaloo ture was neither a necessary nor accurate conversion to
and Hightower Trails (in the vicinity of Etowah) and evaluate relative cost distances. Chi-square analyses
more frequently lie near those specific arteries. Al- compared sites of each temporal affiliation to a set of
though this probably represents an overemphasis on randomly distributed points. Expected values were
survey around the Etowah complex of sites, the absence calculated by extracting the total percentage of land
of Paleoindian and low quantity of Archaic sites in that units in each of the categories and multiplying by the
heavily surveyed area strongly suggests that the total number of sites in each temporal period. If sites
Tugaloo Trail has great time depth (with both Paleo- were randomly distributed, they would fall evenly

84
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Figure 8. Extents of probable Historic trail locations.

across the landscape. The 36 Chi-square evaluations lands, we should see a greater tendency toward locations
were generally significant for each temporal period, farther from primary pathways, as did occur during
each period by phase (where possible), and each period the Archaic.
by primary site type. Of course, the sample size is far Woodland period sites show a tendency to mirror the
too small for Paleoindian or Historic Indian sites to Paleoindian, with a high positive correlation between
consider the results significant; they are shown here sites and areas within 100 cu of the primary pathways,
merely for completeness. Table 2 illustrates the Chi- and a fair negative association between sites and areas
square evaluations for ‘‘all site types’’ only. greater than 100 cu. Unlike the Paleoindian, however,
Results of the statistical analysis indicate a correlation Woodland sites tend to show a greater exploitation of
between archaeological sites (regardless of site func- local resource areas adjacent to the primary pathways.
tion) and primary pathways for most time periods. The In other words, although still a negative correlation, it
strength of the correlation, however, differs between is not as strong as the few Paleoindian sites seem to
periods. Table 3 illustrates another view of the significant suggest.
relationships between sites and distance in cost units by Mississippian sites go one step further with a very
temporal period. By using the proportions of observed high tendency to occur within 100 cu of a primary path-
minus expected sites, we obtain a standardized relative way and a strong negative correlation with areas greater
perspective on the same values shown in Table 2. than 100 cu from a primary pathway. This suggests a
Although the Paleoindian site sample is too small high regard for the travel arteries and little attention to
to be reliable statistically, there is a general trend in adjacent resource areas. A high reliance on agriculture
the Paleoindian site data toward a positive association (which tended to occur in large river valleys, that is,
with primary pathways. Archaic sites only show a weak areas near primary pathways) and trading relationships
positive correlation with locations near primary path- for subsistence might lead to such conditions. In effect,
ways. There is a slight positive correlation with areas the Woodland sites split the difference between the
farther than 300 cu, but it is nearly a random distribu- Archaic and the Mississippian.
tion. This contrast between Paleoindian and Archaic site The Historic Indian sites follow a pattern similar to
relationships with primary pathways seems to suggest that of the Woodland. Although significance is accept-
that while Paleoindian populations traveled through the able at a probability value just over 0.003, the sample
region (probably along what later became known as size does not provide much confidence in the identified
the Tugaloo Trail), they were not exploiting adjacent relationships. A high positive correlation with areas
resource areas to the same degree as the Archaic close to primary pathways is offset by a slight tendency
populations. If they had dispersed their sites in the up- toward utilization of adjacent resource zones. As we

85
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

Table 1. Associations with named primary pathways (Historic period trails).

Number (%) of Sites within 500 Meters, by Period
Trail
Paleoindian Archaic Woodland Mississippi Historic Indian Historic European

Tugaloo 5 (50%) 51 (24%) 53 (32%) 47 (39%) 9 (36%) 51 (14%)

New Echota 0 4 (2%) 6 (4%) 9 (8%) 1 (4%) 5 (1%)
Warwoman’s 0 0 0 0 0 1 (0.2%)
Hightower 0 10 (5%) 9 (5%) 9 (8%) 3 (12%) 12 (3%)
Cherokee Federal Road 1 (10%) 3 (1%) 2 (1%) 0 0 17 (5%)
Total sites 10 215 166 119 25 373

might expect for this time period, a breakdown in resource usage. When compared to secondary path-
trading relationships tended to create a decline in the ways, though, a slightly different picture emerges.
use of primary travel arteries, possibly in favor of more We initially expected that Archaic sites, in particular,
local non-agricultural resource exploitation. In contrast, would be strongly positively correlated with secondary
during the Woodland period, we might have expected pathways since the sites probably represent an exploi-
the development of an extensive trade network to create tation of local upland resources away from primary
the same conditions early on. pathways, a trend reversed by later occupations. In
Historic European sites show the least tendency to other words, we expected a reversal of the results seen
correlate with primary pathways. Although there is a in Tables 2 and 3, with Archaic sites showing the
slight positive association with areas less than 300 cu strongest association with cost distance to secondary
from primary pathways, the tendency to follow known paths and Mississippian sites showing the weakest.
Indian paths and established waterways pretty much Instead, we found that Mississippian sites showed
can account for that. Historic European settlement is not the greatest tendency to occur near secondary path-
well linked to the most effective travel arteries in the ways and Archaic sites showed no significant positive
region. correlation. Woodland sites also showed a significant
association with secondary pathways (Table 4). Given
our current knowledge of the regional distribution of
Temporal Associations with Secondary Pathways
archaeological sites, how should we interpret these seem-
So far, the associations between primary pathways ing contradictions in the associations of settlements with
and sites of different temporal periods are what we both primary and secondary pathways? If sites became
expected, given our current understanding of regional more tightly bound to primary pathways as a result of

Figure 9. Additional primary pathways with long temporal depth.

86
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

Table 2. Chi-square evaluations by period (all site types) for primary pathways.
Relationship between Sites and All Primary Pathways, by Period
Class (cost units) Paleoindian Archaic Woodland Mississippi Historic Indian Historic European
O* E* O E O E O E O E O E

0100 cu 7 3 99 68 118 52 109 37 18 7 131 117

100200 cu 1 2 33 42 10 32 3 23 2 5 97 73
200300 cu 2 1 23 30 13 23 2 17 2 4 62 53
300400 cu 0 1 19 22 9 17 1 12 1 3 38 39
400500 cu 0 1 8 17 3 13 2 9 1 2 23 29
500600 cu 0 1 10 12 2 9 0 6 0 1 12 20
600700 cu 0 1 11 8 6 7 1 5 1 1 6 15
700800 cu 0 0 4 6 1 5 0 3 0 1 2 10
800900 cu 0 0 7 3 2 3 1 2 0 0 1 6
900þ cu 0 0 1 7 2 5 0 3 0 1 1 11
Total sites 10 215 166 119 25 373
Chi-square 10.833 35.469 125.619 201.963 24.919 40.619
P-value 0.287352 ,0.00005 ,0.00005 ,0.00005 0.003070 ,0.00005

* O ¼ observed values, E ¼ expected values.

increased trade, more formal agricultural practices, and environment. As a result, few areas actually fall a long
the movement of populations through the region, why cost distance from the secondary pathways (in fact,
should we see an increase in associations with second- none lie farther than 662 cu), making significance test-
ary pathways as well? The answer may lie in the nature ing of identified associations next to impossible (since
of resource exploitation, the distribution of secondary the expected values are so high). This problem can be
pathways across the environment, and the types of alleviated in the future by addressing specific secondary
archaeological sites identified as ‘‘Archaic.’’ pathways associated with intensively utilized micro-
When Archaic sites are considered by site type (as regions (perhaps the costsheds), which further indicates
recorded on state site forms), all sites identified as the great impact that local-scale analyses may have on
villages or camps occur within 200 cu of a secondary our understanding of large-scale patterns of settlement.
pathway. In fact, all but two fall within 100 cu. Many Regional-level analyses cannot always provide the best
of the sites identified as artifact scatters or quarries, data.
however, occur in areas fairly distant (greater than 200 In contrast, a high positive association between
cu) from secondary pathways, suggesting that they secondary pathways and Woodland, Mississippian,
represent locations of resource collection, not ap- and Historic Indian sites may result from the tendency
proaches to resource zones. This could account for the to locate such sites near major pathway intersections,
low association between Archaic sites in general and both primary and secondary. This is particularly clear
areas near secondary pathways, since so many sites in where many villages are located at the junction of major
the region are merely identified as artifact scatters. In river valleys (the primary pathways through the region)
other words, their dispersion away from all pathways and small tributary streams (nearly always a secondary
reflects the high extent of upland resource utilization in pathway into the surrounding uplands). For instance,
the region during the Archaic. of the 109 Mississippian sites located within 100 cu of
The nature of the topography has also created an even a primary pathway, every one is also located within
dispersal of identified secondary pathways across the 100 cu of a secondary pathway. This has the effect

Table 3. Observed minus expected proportions of sites by temporal period and primary pathways.
Observed-Expected Percentage of Sites, by Temporal Affiliation
Class (cost units)
Paleoindian Archaic Woodland Mississippi Historic Indian Historic European

0100 cu 38.50 14.54 39.58 60.09 40.50 3.62

100200 cu 9.47 4.12 13.44 16.95 11.47 6.54
200300 cu 5.85 3.46 6.32 12.47 6.15 2.47
300400 cu 10.33 1.50 4.91 9.49 6.33 0.15
400500 cu 7.74 4.02 5.93 6.06 3.74 1.57
500600 cu 5.40 0.75 4.19 5.40 5.40 2.18
600700 cu 3.94 1.18 0.32 3.09 0.06 2.33
700800 cu 2.74 0.88 2.13 2.74 2.74 2.20
800900 cu 1.60 1.65 0.40 0.76 1.60 1.33
900þ cu 3.14 2.67 1.93 3.14 3.14 2.87

Notes: Positive correlations are bold/italic; higher values ¼ stronger correlation.

87
SOUTHEASTERN ARCHAEOLOGY 22(1) SUMMER 2003

Table 4. Chi-square evaluations by period (all site types) for secondary pathways.
Relationship between Sites and All Secondary Pathways, by Period
Paleoindian Archaic Woodland Mississippi Historic Indian Historic European
Class (cost units) O* E* O E O E O E O E O E

0100 cu 10 6 152 140 140 106 114 77 23 16 247 242

100200 cu 0 3 48 58 22 45 4 32 2 7 104 100
200300 cu 0 1 13 15 4 12 1 8 0 2 19 26
300400 cu 0 0 2 2 0 2 0 1 0 0 1 4
400þ cu 0 0 0 0 0 0 0 0 0 0 2 0
Total sites 10 215 166 119 25 373
Chi-square 6.67 3.019 29.99 49.40 8.63 4.40
P-value 0.154616 0.554596 ,0.00005 ,0.00005 0.070956 0.354844

* O ¼ observed values, E ¼ expected values.

of simulating a strong utilization of upland resources In a general context, associations between primary
merely by association with the secondary pathways, and secondary pathways and sites of different temporal
when in actuality it is probably an autocorrelation due to periods suggest an ephemeral usage of the region dur-
the cultural selection of terraces near stream intersections ing the Paleoindian, followed by intensive upland re-
as good habitation areas for other reasons, such as access source exploitation during the Archaic. The Woodland
to bottomland resources. period shows an increasing focus on primary pathways,
For the Paleoindian period, the picture is somewhat perhaps as trading relationships with neighboring re-
different. Several sites are located in upland ridgetop gions were established. This intensifies to its greatest
environments, not suggestive of a selection for stream extent during the Mississippian period, when formal
intersections. Instead, we suggest that such sites become agricultural economies tied sites more closely to major
strongly associated with secondary pathways, which waterways and floodplains. The Historic period brought
provide the most efficient ways to climb short adjacent about a transition to fewer associations with primary
ridges that overlook primary or other secondary path- pathways as trading relationships declined.
ways. This is probably as a hunting strategy and seems to This series of analyses explored the data on hand to
follow the pattern identified by O’Steen (1996:97) in the determine how we might extract natural pathways from
Oconee River Valley (just east of the study area), where the known environment and explain their usage in the
the majority of Paleoindian sites were found in the context of an archaeological dataset. Several potential
floodplain with a smaller percentage located in the avenues for future research may prove fruitful. First
uplands immediately adjacent to the river valley. In and foremost, significantly more survey in the region
contrast, an association with upland lithic sources might would provide additional data to support or dispute the
also account for the perceived pattern, much like that assertions made here. This would be especially beneficial
envisioned by Gillam (1996). Once again, though, the in regard to the Paleoindian occupations, but additional
sample size is very small for Paleoindian sites and site information from all periods would enable future
nothing definitive can be addressed without more data. researchers to develop a better understanding of how
different people used primary and secondary pathways.
Second, taking the data to the next level might clarify
Conclusions some of the issues we have discussed. For instance, it
may be possible to develop datasets relating to pre-
GIS evaluation of this part of North Georgia suggests historic patterns of vegetation and wetlands that may
some ideas regarding primary and secondary pathways substantially alter the friction surfaces and require
in the region. The Historic period trails seem to correlate revision of the regional cost topography. Similarly,
well with specific travel arteries identified through least adding the concept of riverine travel to the analysis
cost path analysis. Evaluation of the time depth of such could allow a reinterpretation of the flow accumulation
arteries suggest that the Tugaloo Trail, in particular, surface to focus on additional least cost pathways. Soil
seems to have had a long use, probably dating to the agricultural capacities, slope aspect, and lithic source
Paleoindian period. In contrast, the New Echota and surfaces may all lead to more refined definitions of
Hightower Trails seem to have been popular only dur- travel costs and site attractors.
ing the Woodland and later periods, perhaps serving Third, an evaluation of regional settlement in the
as branches of primary trade networks, specifically for context of cognitive issues is possible by incorporating
the Etowah complex of sites. Few data exist to interpret viewshed reconstructions, recreating territorial bound-
the time depth of the Cherokee Federal Road and the aries, addressing group size dynamics, and assessing
Warwoman’s Trail. patterns of conflict and migration. Such an approach

88
 PREHISTORIC AND HISTORIC TRAVEL CORRIDORS

could inform us on such varied topics as Paleoindian Sweden, edited by G. Burenholt, pp. 5B12. BAR Inter-
hunting strategies, the effects of neighboring group national Series 1016. Archaeopress, Oxford, UK.
pressures on Archaic peoples, the development of trade Binford, Lewis R.
networks during the Woodland, patterns of Mississip- 1989 Debating Archaeology. Academic Press, New York.
Crescioli, Marco, Andrea D’Andrea, and Franco Niccolucci
pian political domination, and the decline of political
2000 A GIS-based Analysis of the Etruscan Cemetery of
superstructures and the out-migration of Historic Indian
Pontecagnano Using Fuzzy Logic. In Beyond the Map:
populations. Archaeology and Spatial Technologies, edited by G. Lock. IOS
Press, Amsterdam, The Netherlands.
Notes Christaller, W.
1935 Die Zentrallen Orte in Sueddeutschland. J. Fischer Verlag,
Acknowledgements. An early version of this paper was Jena, Germany.
presented at the Southeastern Archaeological Conference in Downs, R., and D. Stea
Chattanooga as Evaluating the Potential for Prehistoric Travel 1977 Maps in Minds: Reflections on Cognitive Mapping. Harper
Arteries Across the Northern Arc Study Area in North Georgia Row, New York.
Using GIS, on November 17, 2001. The initial data analysis for Douglas, David H.
the project was undertaken with the support of the Georgia 1994 Least Cost Path in GIS Using an Accumulated Cost
Department of Transportation (GDOT). Grateful acknowledg- Surface and Slope Lines. Cartographica 31(3):37–51.
ment is given to GDOT, and the authors wish to express their Gaffney, Vincent, Zoran Stancic, and Helen Watson
hope that many more researchers in both CRM and academia
1996 Moving from Catchments to Cognition: Tentative Steps
will pursue the use of CRM-generated project data for their
Toward a Larger Archaeological Context for GIS. In
own research.
Anthropology, Space, and Geographic Information Systems,
edited by M. Aldenderfer and H. D. G. Maschner, pp.
References Cited 132–154. Oxford University Press, Oxford, UK.
Gillam, J. Christopher
Aldenderfer, Mark, and Herbert D. G. Maschner (eds.) 1996 Early and Middle Paleoindian Sites in the Northeastern
1996 Anthropology, Space, and Geographic Information Systems. Arkansas Region. In The Paleoindian and Early Archaic
Oxford University Press, Oxford, UK. Southeast, edited by David Anderson and Kenneth Sassa-
Allen, Kathleen M., Stanton E. Green, and Ezra B. W. Zubrow man, pp. 29–57. University of Alabama Press, Tuscaloosa.
(eds.) Gillings, Mark, David Mattingly, and Jan van Dalen (eds.)
1990 Interpreting Space: GIS and Archaeology. Taylor and 1999 Geographical Information Systems and Landscape Archaeol-
Francis, New York. ogy: The Archaeology of Mediterranean Landscapes 3. Alden
Anderson, David G., and J. Christopher Gillam Press, Oxford, UK.
1997 Paleoindian Colonization of the New World: Implica- Hargrett Rare Map Collection
tions from an Examination of Physiography, Artifact 1779 A New and Accurate Map of the Province of Georgia in
Distributions, and Demography. Paper presented at the North America. Hargrett Rare Map Collection, University
62d annual SAA Conference, Nashville, TN. of Georgia Library, Map #1779:N42.
2000 Paleoindian Colonization of the Americas: Implications Hartley, Ralph, and Anne M. Wolley Vawser
from an Examination of Physiography, Demography, and 1998 Spatial Behavior and Learning in the Prehistoric
Artifact Distributions. American Antiquity 65(1):43–66. Environment of the Colorado River Drainage (Southeast-
Barcelo, Juan A., and J. M. Faura ern Utah), Western, North America. In The Archaeology of
1997 Time Series and Neural Networks in Archaeological Rock Art, edited by C. Chippendale and P. S. C. Taçon, pp.
Seriation: An Example on Early Pottery from the Near 185–211. Cambridge University Press, Cambridge, UK.
East. In Computer Applications and Quantitative Methods Hemperley, Marion
in Archaeology 1997. BAR International Series. Oxford, UK. 1978 Major Indian Trails: Original Research and Sketch
Bell, Suzanne Map. Office of the Secretary of State, Surveyor General’s
1998 Artificial Neural Networks as a Tool for Archaeological Department, Atlanta, GA.
Data Analysis. Archaeometry 40(1):139–151. Hirtle, Stephen, and Alan M. MacEachren (eds.)
Bell, Suzanne, and Richard Jantz 1998 Cognitive Models of Dynamic Geographic Phenomena and
2002 Neural Network Classification of Skeletal Remains. Their Representations. Varenius Workshop Report, Pitts-
In Archaeological Informatics: Pushing the Envelope. burgh, PA, October 28–31, 1998.
Proceedings of the 2001 Computer Applications and Hodder, Ian
Quantitative Methods in Archaeology Conference, Visby, 1986 Reading the Past: Current Approaches to Interpretation in
Sweden, edited by G. Burenholt, pp. 205–211. BAR Inter- Archaeology. Cambridge University Press, Cambridge, UK.
national Series 1016. Archaeopress, Oxford, UK. Hodler, Thomas W., and Howard A. Schretter
Bellavia, Gino 1986 The Atlas of Georgia. University of Georgia Press, Athens.
2002 Extracting ‘‘Natural Pathways’’ from a Digital Elevation Jensen, Dana E.
Model: Applications to Landscape Archaeological Studies. 2002 Geoglyphs and GIS: Using GIS to Model Transhumance
In Archaeological Informatics: Pushing the Envelope. in the formative of Northern Chile. Paper presented at the
Proceedings of the 2001 Computer Applications and Computer Applications in Archaeology 2002 Conference,
Quantitative Methods in Archaeology Conference, Visby, Herkalion, Crete, Greece, April 2–6, 2002.

89
Langren, Gail Swanton, John
1992 Time in Geographic Information Systems. Taylor and 1922 Early History of the Creek Indians and Their Neighbors.
Francis, London, UK. Bureau of American Ethnology Bulletin 73. Smithsonian
Lindgren, Ernesto S. Institution, Washington DC.
1967 Proposed Solution for the Minimum Math Problem. Tobler, W.
Harvard Papers in Theoretical Geography: Geography and the 1993 Non-isotropic Geographic Modeling. In Three Presen-
Properties of Surfaces Series 4. Cambridge University Press, tations on Geographic Analysis and Modeling. National
Cambridge, UK. Center for Geographic Information and Analysis, University of
Lock, Gary (ed.) California Technical Report 93-1. Santa Barbara, CA.
2000 Beyond the Map: Archaeology and Spatial Technologies. IOS Warntz, William
Press, Amsterdam, The Netherlands. 1957 Transportation, Social Physics, and the Laws of Re-
Lock, Gary, and Zoran Stancic (ed.) fraction. Professional Geographer 9(4):2–7.
1995 Archaeology and Geographic Information Systems: A 1965 A Note on Surfaces and Paths and Applications to Geo-
European Perspective. Taylor and Francis, London, UK. graphical Problems. Michigan Inter-University community
Maschner, Herbert D. G. of Mathematical Geographers, Ann Arbor.
1996 The Politics of Settlement Choice on the Northwest Warren, Robert E.
Coast: Cognition, GIS, and Coastal Landscapes. In 1990 Predictive Modeling of Archaeological Site Location: A
Anthropology, Space, and Geographic Information Systems, Case Study in the Midwest. In Interpreting Space: GIS and
edited by Mark Aldenderfer and H. D. G. Maschner, pp. Archaeology, edited by K. M. Allen, S. W. Green, and E. B. W.
175–189. Oxford University Press, Oxford, UK. Zubrow, pp. 201–215. Taylor and Francis, New York.
Mitchell, John Weber, Alfred
[1755] The Southeastern Part of the Present United States. 1929 Theory of the Location of Industries. University of Chicago
In Early History of the Creek Indians and Their Neigh- Press, Chicago, IL.
bors. J. Swanton (1922). Bureau of American Ethnology Wescott, Konnie L., and R. Joe Brandon (eds.)
Bulletin 73. Smithsonian Institution, Washington, DC. 2000 Practical Applications of GIS for Archaeologists: A Predictive
Niccolucci, Franco, Andrea D’Andrea, and Marco Crescioli Modeling Kit. Taylor and Francis, London, UK.
2001 Archaeological Applications of Fuzzy Databases. In Wescott, Konnie L., and James A. Kuiper
Computing Archaeology for Understanding the Past: 2000 Using a GIS to Model Prehistoric Site Distributions in
CAA 2000, Proceedings of the 28th Conference, Ljubljana, the Upper Chesapeake Bay. In Practical Applications of GIS
April 2000, edited by Z. Stancic and T. Veljanovski, pp. for Archaeologists: A Predictive Modeling Kit, edited by K.
107–116. BAR International Series 931, Archaeopress, Wescott and J. Brandon. Taylor and Francis, London, UK.
Oxford, UK. Wheatley, David
O’Steen, Lisa 1993 Going Over Old Ground: GIS, Archaeological Theory
1996 Paleoindian and Early Archaic Settlement Along the and the Act of Perception. In Computer Applications and
Oconee Drainage. In The Paleoindian and Early Archaic Quantitative Methods in Archaeology, CAA ’92, edited by J.
Southeast, edited by David Anderson and Kenneth Sassa- Andresen, T. Madsen, and I. Scollar, pp. 133–138. Aarhus
man, pp. 92–106. University of Alabama Press, Tuscaloosa. University Press.
Purcell, Joseph Whitley, Thomas G.
[1770] Indian Nations . . . compiled not later than 1770 in 2000 Dynamical Systems Modeling in Archaeology: A GIS
the interest of British Indian Trade by John Stuart, His Approach to Site Selection Processes in the Greater Yellow-
Majesty’s Superintendent of Indian Affairs. In Early His- stone Region. Ph.D. dissertation, Department of Anthro-
tory of the Creek Indians and Their Neighbors. J. Swanton pology, University of Pittsburgh. University Microfilms,
(1922). Bureau of American Ethnology Bulletin 73. Smithso- Ann Arbor, MI.
nian Institution, Washington, DC. 2002 Modeling Archaeological and Historical Cognitive
Reeler, C. Landscapes in the Greater Yellowstone Region Using
1999 Neural Networks and Fuzzy Logic Analysis in Archae- Geographic Information Systems. In Archaeological In-
ology. In Computer Applications and Quantitative Methods formatics: Pushing the Envelope. Proceedings of the 2001
in Archaeology 1998. BAR International Series 750 (supple- Computer Applications in Archaeology Conference, Visby,
mentary). Oxford, UK. Sweden, edited by G. Burenholt, pp. 139–148. BAR
Renfrew, Colin International Series 1016. Archaeopress, Oxford, UK.
1994 Towards a Cognitive Archaeology. In The Ancient Mind: Witcher, Robert
Elements of a Cognitive Archaeology, edited by Colin Renfrew 1999 GIS and Landscapes of Perception. In Geographical Infor-
and Ezra Zubrow, pp. 3–12. Cambridge University Press, mation Systems and Landscape Archaeology: The Archaeology of
Cambridge, UK. Mediterranean Landscapes 3, edited by M. Gillings, D. Matt-
Salmon, Wesley C. ingly and J. van Dalen, pp. 13–22. Alden Press, Oxford, UK.
1971 Statistical Explanation and Statistical Relevance. University Zubrow, Ezra B. W.
of Pittsburgh Press, Pittsburgh, PA. 1994 Knowledge Representation and Archaeology: A Cogni-
1984 Scientific Explanation and the Causal Structure of the World. tive Example Using GIS. In The Ancient Mind: Elements of a
Princeton University Press, Princeton, NJ. Cognitive Archaeology, edited by Colin Renfrew and Ezra
1998 Causality and Explanation. Oxford University Press, Zubrow, pp. 107–118. Cambridge University Press, Cam-
Oxford, UK. bridge, UK.

90

View publication stats