W.W. Baden
Department of Anthropology
University of Tennessee
Knoxville,TN 37996

From: Exploring Tennessee Prehistory: A Dedication to Alfred K. Guthe, edited by T.R. Whyte, C.C. Boyd, Jr., and B.H. Riggs. The University of Tennessee, Department of Anthropology, Report of Investigations 42


Analysis of data from the 1979-82 probabilistic survey of the Tellico Reservoir in east Tennessee demonstrates that surface collections can be used to generate a model of settlement change over time. Steady increases in site density along major river valleys are shown to have occurred between 7900-200 B.C.. During the Middle Woodland period (200 B.C.-A.D. 900) intensive site utilization shifted to the mouths of primary streams. The Mississippian settlement pattern is marked by the extensive expansion onto cultivatable soils.


A three phase probabilistic survey of east Tennessee's Tellico Reservoir was undertaken between 1979 and 1982 (Davis et al. 1982). Each phase involved the independent sampling of three areas within the study region: 1) the below pool river terraces, 2) the Tennessee Valley Authority's designated Industrial Area II (Davis 1980), and 3) the above pool uplands. The primary goal of this project was the generation of a predictive site location model for aboriginal occupations that could serve as a cultural resource management tool for the Tennessee Valley Authority. To produce this model, data was gathered from controlled surface collections of 425 300 x 300 ft randomly selected sample units (SU's) of which 264 produced aboriginal material. These units were drawn from a population of 12,663 units that covered 26,212 acres of the T.V.A. Tellico property. An additional 8232 acres of forested land was excluded from the sample because they could not be comparably collected [this exclusion produced only minor biases in the overall sampling scheme (Davis 1982a)].

The original predictive model did not address differential site location through time. This paper will examine certain temporal patterns in the sample. In particular, I will examine the evidence for changing aboriginal utilization of the study area over a period of 9000 years. It will be demonstrated that systematic sampling, when carried out over large contiguous areas, can provide interpretable evidence of settlement variability across time.


The 1979-82 survey data consist of aboriginal lithic and ceramic remains recovered from 425 SU's randomly located within the bounded T.V.A. property lines of the Tellico Reservoir. Each SU was composed of six 20 x 300 ft plowed transects that divided the unit into nine cross intersections of three equally spaced north-south transects and three east-west strips ( Figure 1). This pattern represents a 36 percent (0.74 acres) systematic sample of the entire unit.

The entire 425 unit sample was derived from three independent random selection schemes. The first sample stratum consisted of 100 units located in the below pool portion of the reservoir area. Its sampling fraction, f, was 0.0151. The second stratum was the T.V.A. designated upland Industrial Area II. Its 0.05 sample fraction contributed 64 units to the overall sample. The final stratum consisted of 261 units (f=0.0553) taken from the remaining above pool portion of the study area.

It was determined (Baden 1982a) that a minimum of two to three inches of rainfall exposure was required to insure an adequate and complete surface collection following systematic plowing to depths generally between .8-1.0 ft. In a limited number of cases this requirement meant that units had to be recollected or artificially watered using a trash pump and water hose. This collection strategy has been shown (Baden 1982a) to produce a plowzone sample fraction along a gradient (0.029 < f < 0.123) directly proportional to individual artifact size. Although the overall small sample fraction precludes determination of site significance in any traditional sense, the SU artifact composition can serve as a first approximation of the temporal and functional characteristics of each area.

Given this collection design, a probabilistic site location model was produced (Baden 1982b). The total sample was divided into four site types defined by pattern recognition techniques that identified similar SU's on the basis of assemblage content (Davis 1982c). (The use of assemblage in this context refers only to the surface sample as a stochastic collective unit and is not intended to convey any implied evaluation of the completeness of the collection as a cultural or temporal composite). These types were designated nil sites (no material recovered), residential bases, field camps, and locations following, in part, Binford's (1982) functional schema for sites (Davis 1982c). It was possible to differentially correlate site types with spatial loci over the entire study area by topographically defining the units in terms of the quantifiable variables of elevation above the nearest water source, elevation above mean sea level, distance to the nearest major stream (Tennessee, Little Tennessee, or Tellico rivers), and distance to the nearest primary tributary (Fork, Notchy, Bat, Citico, Island, Baker, Ninemile, or Ballplay creeks). The result was the generation of four probabilistic maps locating areas most likely to yield each of the site types (see Figures 2, 3, 4, and 5).

Although this approach facilitates the location of potentially significant resources thus helping, to insure their protection, it fails to address settlement system changes and redundancies through time. Failure to adequately account for temporal dependencies somewhat distorts the true picture of aboriginal settlement in the lower Little Tennessee River valley. Lithic (Kimball 1982) and ceramic (Davis 1982b) temporal correlates were associated with 96 SU's and ten buried contexts in T-1 deposits along the Little Tennessee River. Five temporal units (or components) were recognized in terms of artifact pattern recognition models applied to debitage, formalized tools, and sherd assemblages: 1) Early Archaic (7900-6100 B.C.), 2) Middle Archaic (6000-4300 B.C.), 3) Late Archaic/Woodland 1 (3000-200 B.C.), 4) Woodland 11/111 (200 B.C.-A.D. 900), and 5) Mississippian (A.D. 900-1819).

As an example of the importance of recognizing diachronic effects, had temporal relationships been used to evaluate the 'functional' site types that helped define the site location model, it would have been clear that the artifact patterns more appropriately measured the reuse of loci over time rather than site function. Table 1 presents the crosstabulation of the number of components recognized per SU versus the associated site type for the 264 units producing aboriginal material. Units producing material but lacking any temporal diagnostics tend (p=0.6369) to be defined as locations. Units with only one recognized temporal unit tend to be called either field camps (p=0.4286) or locations (p=0.3469). If the unit had two or more components present it tended to be classified as a residential base (p=0.8511). Clearly the original cluster designations are more useful as measures of repeated use rather than as categories of site function. Predictive modelling of site location can be improved by considering temporal parameters even though the useable sample size is significantly smaller. The following discussion will take this sample and examine its spatial characteristics in terms of the probabilistic implications for identifying trends in site use and reuse over time.

Table 1. Distribution of sample units in comparison to the number of recognized temporal components and classified functional site designations.

Number of dated
Residential BasesField CampsLocationsTotal



To develop a model of land use over time we must begin by identifying, based on the characteristics of the random sample, sets of spatial loci which display differential potential for human habitation and assign them temporally dependent probabilities of occupation. If these spatial loci or topographic types can be quantitatively defined in a manner that permits the classification of any portion of the study area, then a site prediction model can be produced using spatial interpolation procedures. This is accomplished by applying basic pattern recognition techniques to the SU's defined in terms of the four topographic variables.

The first step of this process involves standardizing and simplifying the four variables by applying principal component analysis with varimax rotation to the correlation matrix derived from the 96 datable units. The resulting solution provides dual components which scale the elevation and distance to water variables independently. The two component characterization of the 96 dated units can be used to define minimum variance clusters following Ward's algorithm (Ward 1963). An examination of the fluctuations in the clustering criterion (Figure 6) indicates either a three or nine cluster solution as the best fit for this data. The nine cluster solution was preferred because the smaller grouping tended to oversimplify the topographic variability into merely river terraces, slopes, and uplands. A discriminant analysis incorporating the nine within group covariance matrices derived from the clustered 96 datable SU's can then be used to assign locational type designations to each of the 329 remaining units. (The within group covariance matrices were used rather than the pooled form because the between group variances were not equal). A breakdown of these nine quantitatively defined clusters in terms of nine recognizable land form types is given in Table 2. Table 3 provides mean and standard deviations for the original four variables within each cluster. The clusters appear to reflect the topographic gradient that separates river terraces from the uplands across the study area.

Table 2. Distribution of cluster groups by major land form types.

Cluster Group
Land form123456789

T-1 LTR15104020000
T-2 LTR811000000
Older Terraces LTR491055120
LTR slopes0081412080
Tellico R. valley485000000
Tellico R. slopes111202020
Prim. trib. valley04001301300
Prim. trib. slopes0104123900

LTR = Little Tennessee River
Prim. trib. = Primary tributary

Table 3. Statistics on original topographic variables by cluster group.


137852.04953.489924.723561.5710.658.70 798.7518.99
249 1281.22 1457.374084.733443.0212.859.02 795-3413.40
321 600.91570.93 22979.918181.8520.5919.01 805.5531.05
4128 3153.09 2390.315983.745479.08106.0453.90 915.9656.51
556 4489.32 1274.791988.352508.51 15.059.91815.0019.35
640 1894.32 1232.084763.643333.91 35.1715.25832.6819.15
745 7078.58 4073.855082.606758.22152.04 184.71972.69 180.98
837 1420.00 1092.23 21198.21 8347.92 63.27 32.72 859.4029.42
912 1864.24 1197.80 7922.73 6178.87 133.33 20.46 921.0625.14

C = Cluster group
DIST1 = Distance to the nearest major stream (ft)
DIST2 = Distance to the nearest secondary stream (ft)
ELEV1 = Elevation above the nearest water source (ft)
ELEV2 = Elevation above mean sea level (ft)

To produce the original regional maps (Figures 2, 3, 4, and 5), 915 additional unsampled SU locations were quantified in terms of the four topographic variables. For this study component scores are calculated and the discriminant analysis is used to assign locational type designations for each. The combined 1340 unit sample can now be used to define the spatial distribution of each of the nine location types across the study area. By associating probabilities of site occurrence with each cluster group, maps that delineate levels of site concentration for each time period can be interpolated using the piecewise Bessel algorithm of SURFACE II (Sampson 1978).

Calculation of the relative frequency of each temporal component within each location type can be used as an estimate of the probabilities of site occurrence. However, because separate sampling schemes were used to select the 425 SU's, each unit must be weighted by the inverse of its associated sample fraction in order to produce an unbiased estimate of the total population density of each cluster type. Thus, one unit from the below pool sample stratum (f=0.0151) would represent 66.23 SU's from the total population. Units from the second and third strata would correspond with 20.00 and 18.08 units, respectively. The final weighted estimates of the frequency of location types for the 12,663 unit population are:

1 2 3 4 5 6 7 8 9
1969.81 2524.03 716.94 2434.9 2094.93 832.06 1152.67 717.14 219.46

The expected relative frequency of each temporal component for each location type is provided in Table 4. Each value corresponds with the probability of locating a specific component within a 300 x 300 ft area sampled in the manner described above. These values can be interpreted as the expected density of temporally recognizable sites. To graphically present the shifts in settlement density and location, maps were produced showing the differences between contiguous temporal units (Figures 7, 8, 9, and 10). (Areas not highlighted represent consistent or decreased levels of utilization between time periods and not an expected absence of sites). This provides the basis for an evaluation of settlement changes through time.

Table 4. Probabilities of site location on nine topographic types by temporal period.

Topographic Types

Early Archaic.
Middle Archaic.
Late Archaic/
Woodland I
Woodland II/III.388.


The approach applied to this data is one of crosstabulation of categorical data to produce estimated probabilities of event occurrence at any given locus. Because the units were randomly selected, each temporal period has a somewhat equal a priori chance of being identified at any one location. Obviously the probability of identifying a component is dependent on the number of diagnostic types associated with each time unit and the total duration of the period. Unfortunately, it is beyond the scope of this paper to fully address these biases. However, this data can be confidently used to identify cultural preferences (trends) for site location, or more explicitly, to identify locations which tend to be correlated with the preservation of temporal markers. Higher probabilities should be directly related to extensive use of loci with lower values indicating less utilization. With this in mind several settlement observations can be made from the data's temporal distribution across space.

Given a standard set of a priori notions of southeastern archaeological cultures we would expect to see an uninterrupted increase in site density from Early Archaic through Mississippian along the terraces of the Lower Little Tennessee River valley. This is based on our assumptions of: 1) continuous occupation of the valley over time, 2) an exponential population growth for its inhabitants, and 3) through time, an increased dependence on plant domesticates grown on river bottom soils. A recent examination (Cridlebaugh 1984) of the effects of aboriginal settlement on the area's vegetation pattern based on paleoethnobotanical and pollen samples supports these contentions. Summarizing her work, prior to the Late Archaic/Woodland I period only minor alterations to the landscape were induced by cultural activities. At about 4000 years B.P. the expanded use of cultigens resulted in the initialization of river terrace clearance. As horticulture became a primary means of food production, populations began to concentrate on soils as a resource. Thus, utilization of more and more river terrace land became necessary to support larger and larger populations. Our changing settlement model should reflect these trends if survey data is to be excepted as a useful source of information.

In general our expectations are supported by the maps showing the areas of increased site density. Figure 7 shows a broad based utilization of site locations across the valley above Fork Greek by Middle Archaic times. The Late Archaic/Woodland I site densities suggest a slight expansion of this pattern with particular emphasis placed on broadening the use of river terraces (Figure 8).

By Woodland II/III (Middle Woodland) times, however, this steady expansion has halted (Figure 9). Based on the survey data, only specific river terrace locations are expected to produce indications of expanded site utilization. It is tempting to suggest that land adjoining the mouths of major streams provide the primary locations for intensive site location. Indeed, the two most extensively excavated Middle Woodland sites from the valley are located near the mouth of Tellico River, Icehouse Bottom (40MR23, Chapman 1973, Cridlebaugh 1981) and Patrick (40MR40, Schroedl 1978). Additionally, pollen data from Tuskegee Pond, located near this area, suggest that steady Zea mays production occurred during this period (Cridlebaugh 1984:143). Given the relatively short 1100 year time span the observance of such a concentrated settlement plan is possible, although not expected. What this map reflects may be the less mobile nature of the Middle Woodland populations over a short period of time which contrasts with the pattern produced by the highly mobile settlement system of the previous 6000 years. This shift marks a major change in settlement strategy and probably social organization,. that culminates in the more intensive occupations of the Mississippian period.

The Mississippian pattern (Figure 10) appears to be an expansion of this earlier concentration on river terraces with the exclusion of those areas most intensively used during the preceding Woodland II/III period. Unlike the pre-Woodland II/III expansions, this pattern is more narrowly confined to the prime agricultural soils and develops over a shorter (900 years) period. The observed increased use of these soils logically follows our expectations given the assumed population levels during this period, the increased dependence on Zea mays by these populations, and the eventual dispersed settlement pattern of the later historic Cherokee.


This brief examination into the changing settlement patterns for the Lower Little Tennessee River valley has been useful for three reasons. First, it demonstrates that survey data can be used to assess settlement changes over time. Second, the results can help direct future research by identifying anomalous situations, like that associated with the Middle Woodland Settlements, that cannot be explained by our current unilineal model for culture change. Finally, the settlement model, reduced to a simple graphic format, has been shown to be a tool for interpretation that readily supplements the results of other studies.

Surveys of this sort provide an important avenue for future research. They alone synthesize various analytical results to produce a predictive model. By striving for a simple representation of the model a useful interpretational tool can be produced, despite the complexity involved in construction.

Acknowledgments. This project is an extension of the original survey funded by the Tennessee Valley Authority. I would like to thank Maxwell Ramsey and Bennett Graham of the T.V.A. for their support and my co-workers, Larry R. Kimball and R.P. Stephen Davis, Jr. for their contributions to the production of the original data used here. I would also like to thank Thomas Whyte, Cliff Boyd, and Brett Riggs for their comments and suggestions which greatly improved the contents of this paper. However, I accept full responsibility for the conclusions presented herein. Financial support for the computational aspects of this research was provided by the University of Tennessee's Computing Center and the Department of Anthropology.


Baden, W.W.
1982a The Nature of Surface Collections. In An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley, edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 105-119. Report submitted to the Tennessee Valley Authority.

1982b Appendix I: Generation of a Probabilistic Model of Site Location. In An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 517-524. Report submitted to the Tennessee Valley Authority.

Binford, L.R.
1982 The Archaeology of Place. Journal of Anthropological Archaeology 1:5-31.

Chapman, J.
1973 The Icehouse Bottom Site, 40MR23. Report of Investigations No. 13. Department of Anthropology, University of Tennessee, Knoxville.

Cridlebaugh, P.A.
1981 The Icehouse Bottom Site (40MR23) 1977 Excavations. Report of Investigations No. 35. Department of Anthropology, University of Tennessee, Knoxville.

1984 American Indian and Euro-American Impact Upon Holocene Vegetation in the Lower Little Tennessee River Valley, East Tennessee. Unpublished Ph.D. dissertation, Department of Anthropology, University of Tennessee, Knoxville.

Davis, R.P.S., Jr.
1980 A Preliminary Report of Probabilistic and Nonprobabilistic Archeological Sampling in Industrial Area II Tellico Reservoir, Tennessee. Tellico Archeological Survey Report No. 2. Department of Anthropology, University of Tennessee, Knoxville.

1982a Sampling Design. In An Archeological Survey and Assessment of Aboriginal Settlement within the Lower Little Tennessee River Valley, edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 37-72. Report submitted to the Tennessee Valley Authority.

1982b Ceramic Artifact Analysis. In An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley, edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 253-332. Report submitted to the Tennessee Valley Authority.

Davis, R.P.S., Jr.
1982c Archeological Settlement Analysis. In An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley, edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 333-411. Report submitted to the Tennessee Valley Authority.

Davis, R.P.S., Jr., L.R. Kimball, and W.W. Baden (editors)
1982 An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley. Report submitted to the Tennessee Valley Authority.

Kimball, L.R.
1982 Lithic Artifact Analysis. In An Archeological Survey and Assessment of Aboriginal Settlement Within the Lower Little Tennessee River Valley, edited by R.P. Stephen Davis, Jr., Larry R. Kimball, and William W. Baden, pp. 120-252. Report submitted to the Tennessee Valley Authority.

Sampson, R.J.
1978 Surface II Graphics System. Kansas Geological Survey, Lawrence.

Schroedl, G.F.
1978 The Patrick Site (40MR40), Tellico Reservoir, Tennessee. Report of Investigations No. 25. Department of Anthropology, University of Tennessee, Knoxville.

Ward, J.H., Jr.
1963 Hierarchical Grouping to Optimize an Objective Function. Journal of the American Statistical Association 58:236-244.