Mn/Model Final Report Chapter 8: Model Results and Interpretations

Quick Links

Chapters

Appendices

Chapter 8

Model Results and Interpretations

By Elizabeth Hobbs, Craig M. Johnson, Guy E. Gibbon, Carol Sersland, Mark Ellis, and Tatiana Nawrocki

Statewide Survey Impelmentation Model Map

Chapter 8 Table of Contents
8.1   Introduction
8.2 Model Description
         8.2.1 Environmental Context
         8.2.2 Types of Models
8.3   Model Evaluation
8.4   Model Interpretation
         8.4.1 Presentation of Interpretations
         8.4.2 Previously Identified Variables
8.5   Model Comparison
         8.5.1 Subsection Group Approach
         8.5.2 Site Catchment Analysis
         8.5.3 Cultural Context and Site Location
         8.5.4 SHPO Intuitive Model
8.6   Model Results
         8.6.1 Phase 1 Results
         8.6.2 Phase 2 Results
         8.6.3 Phase 3 Results
8.7   Agassiz Lowlands
8.8   Anoka Sand Plain
8.9   Aspen Parklands
8.10 Big Woods
8.11 Blufflands
8.12 Border Lakes
8.13 Chippewa Plains
8.14 Coteau Moraines / Inner Coteau
8.15 Glacial Lake Superior Plain/Northshore Highlands/ Nashwauk Uplands
8.16 Hardwood Hills
8.17 Laurentian Highlands
8.18 Littlefork-Vermilion Uplands
8.19 Mille Lacs Uplands
8.20 Minnesota River Prairie
8.21 Oak Savanna
8.22 Pine Moraines & Outwash Plains
8.23 Red River Prairie
8.24 Rochester Plateau
8.25 St. Croix Moraines and Outwash Plains (Twin Cities Highlands)
8.26 St. Louis Moraines/ Tamarack Lowlands
8.27 Conclusion
        References

8.1 INTRODUCTION

Chapter 8 presents summaries of the modeling results of Phases 1 and 2 of the project, as well as detailed evaluations of the best models derived in Phase 3. Preliminary sections explain the reporting conventions used for model evaluations, for model descriptions, and the methods used for interpreting the relationships of the model variables and archaeological resources. The final section provides a summary of the Phase 3 model results.

Model development began in the fall of 1995 with a pilot model developed for Nicollet County. After experimenting with various modeling procedures and variables, these were refined and then applied, in fall of 1996, to the Phase 1 counties in each of five regions. Refinements of the Phase 1 procedures lead to the Phase 2 procedures, which were applied statewide in the spring and summer of 1997. These models were delivered in the fall of 1997 and were documented in an interim report in winter of 1997. Following a brief period of data preparation and testing of some new procedures, Phase 3 models were started in early 1998. Phase 3 models were completed in the summer of 1998 and preparation of final project deliverables began in the fall. The design of the GIS to support this model (Chapter 4), the development of the archaeological database (Chapter 5), the environmental variables and archaeological variables (Chapter 6), and the modeling methods (Chapter 7) are all discussed in detail elsewhere in this report.

The models presented are based on Ecological Classification System (ECS) subsections or combinations of subsections (Section 7.5.1.3.4) (Figure 8.1). Separate models were developed for each of 24 subsections. An additional 12 combinations of subsections were modeled to seek better models for subsections with low site numbers. Hence the term region, when used in Phase 3, refers to a geographic modeling unit that may consist of an ECS subsection or set of subsections. The best models for each subsection were selected for implementation (Section 7.5.1.3.4). The models are for 17 individual subsections and for three combined subsections (Section 8.6.4). For six of the individual subsections reported, models are taken from combined subsections. Only when two or more subsections share the same best site probability and survey probability models are they reported on in combination, to avoid redundancy.

Three types of models were developed for each modeled region (Section 7.2). The site probability models considered all site types (except single artifacts) as being either present or absent. The survey probability models considered known sites of all types and negative survey points to represent surveyed places. Site and survey absences were represented by random points distributed throughout the modeled region. For each of these model types, pairs of preliminary models were developed for each region using randomly selected halves of the database. The results of these preliminary model pairs are reflected in the Kappa statistics used to evaluate model stability (Section 7.5.1.3). The final models presented in Section 8.6, however, were built using the entire database (Section 7.5.1.3.2).

The statewide models (Section 8.6.3) are all composites of the regional models, with the exception of the model developed from the statewide database (Section 8.6.3.4). Only brief descriptions and evaluations of these models are presented to provide a view of how well the models work at a statewide scale.

Return to Top

8.2 MODEL DESCRIPTION

Sections 8.2 through 8.4.1 are intended as a guide when reading the regional model reports in Sections 8.7 through 8.26.

8.2.1 Environmental Context

The first element of the regional model reports is a description of the environmental context of the models. The ECS subsection or subsections that make up the modeling region are described with respect to:

How the region is defined and the features that make up its boundary
Geomorphology bedrock geology
Surface hydrology
Topography
Soils
Climate
Historic Vegetation

Two pages of maps per region support these descriptions. One illustrates the physical features and vegetation of the region. The other depicts the geology and soils. All of these represent data layers that were used to derive the environmental variables used for modeling. Physical features include elevation, watersheds, and surface hydrology. Elevation units are feet. Elevation source scales vary, depending on the source (Section 4.5.1.3). Watershed boundaries were mapped by Mn/DNR at a source scale of 1:100,000. Streams were digitized by MnDOT from 1:24,000 USGS quadrangles. Lakes, wetlands and major rivers were taken from the National Wetlands Inventory, which also has a source scale of 1:24,000. The vegetation depicted is that mapped by Marschner (1974) from the PLSS survey records. The source scale of this map is 1:500,000. Surface (quaternary) geology is mapped from the MGC100 dataset from LMIC. The original source was the State Soil Atlas, which was mapped at a resolution of 40 acres. The generalized soils map is from the same source.

8.2.2 Types of Models

Within the regional model discussions (Section 8.6.1) are separate sections for the site and survey probability models. A description of the spatial distribution of the three site or survey potential zones is provided for each, along with a map of the model. Model maps illustrate the geographic pattern of the three potential zones, the distribution of the population of sites used to build the model, and the distribution of a small group of more recently recorded sites that were not available when the archaeological database was assembled. For reference, the maps also show the locations of mines, water, steep slopes, streams, and county boundaries. In the map title bar, below the year, is the specific name of the model depicted. The naming convention for models based on individual subsections has three elements:

The type of model (T for site probability models, V for survey probability models)
The sequence number of the model. In each S-Plus model run, five rank ordered models are provided as output (Section 7.3.6.2). Usually the first in the sequence has a significantly higher posterior probability than the others and is selected. Occasionally, there is very little difference between the first and second models. In these cases, both were applied to the landscape, their performance evaluated. The one that performed best was selected for implementation.
The number of site or survey potential classes. Only models with three site or survey potential classes are presented in this report. Preliminary models with 20 classes used to derive the final models (Section 7.5.1.3.1) are not included.

Model names for combined subsections contain only two elements:

Type of model (T or V, as above)
A four-letter abbreviation of the names of the subsections that were combined to create the model.

In the model descriptions, prominent model patterns are related to the dominant and most easily identified landscape feature – surface hydrology. Where relevant and identifiable, spatial relationships between site or survey potential zones and terrain, geomorphic, or vegetation features are also noted. Specific features, such as lakes, rivers, streams, and county boundaries are identified to more fully integrate the description into the surrounding landscape and to provide reference that the scale of mapping in this report cannot. Survey implementation models were not described because they are a composite of the site and survey probability models and, as such, correspond to the descriptions of these models.

Return to Top

8.3 MODEL EVALUATION

Model evaluation procedures are discussed in Section 7.5. This section focuses on how the results of those evaluations are presented in this report (Section 8.6).

Each regional model discussion provides a summary of the model's performance with respect to the project's goal. This goal has been to predict at least 85 percent of known sites, with 33 percent or less of the landscape classified as high and medium site potential. Achieving the goal should result in a gain statistic of 0.6118 or better (Section 7.5.1.2.3). Model statistics reported include:

The number and general types of model variables.
The percentage of known sites detected by the model.
The percentage of land area classified as high and medium potential.
The gain statistic, a measure of model performance. Values range from zero (no gain over chance alone) to one (a perfect model). This statistic is interpreted as the model’s gain, in terms of the percentage of correct predictions, over chance (Kvamme, 1988, p.329). It is a function of the percent of total area in high and medium potential classes and the percent of known sites within these areas (Section 7.5.1.2.3).
The number of sites in the database used to build and test the model. Site numbers have a strong impact on model reliability and performance. Models built with fewer than about 250 sites are probably suspect (Section 7.5.1.3.4).
The percent agreement between the two preliminary models based on different halves of the same database. This measure is defined as the percentage of cells in the region classified the same in both preliminary models (Section 7.5.1.3.2). It is a precursor to the derivation of the Kappa coefficient, but because it is not adjusted by expected agreement it is usually an overestimate of correlation between the two models.
The Kappa coefficient of agreement (Bonham-Carter, 1994:245), which measures model stability. Stable models are those that show little change no matter which portion of the database is modeled. Kappa is calculated from the percent agreement between the two models, but adjusts this measure to account for the amount of agreement expected due to chance alone (Section 7.5.1.3.2). Values range from negative one (perfect disagreement between the models) to one (perfect agreement). The value zero indicates the amount of agreement expected due to chance.

Within the regional model report, three tables and several maps support the evaluation section. The first table reports the names of the model variables, their regression coefficients, and the variable's probability, reported by S-Plus. No conclusions should be drawn from the magnitude of the regression coefficients or even whether they are positive or negative. Environmental attributes can interact in complex ways so, from these coefficients alone, there is no means of evaluating which variables have the strongest influence on the model. The probability values are better indicators of the relative importance of variables within a model. This value is the probability that the coefficient for a given variable is not zero. Values are given in percentages (0 - 100).

Maps of some model variables are included to illustrate the geographic pattern of the variable. These can be compared to the model maps to gain a visual impression of the correspondence between the variables and the model.

The second table presents the numeric evaluation of model performance. This table reports:

The number of grid cells in each model class and the percentage of those cells as a part of the total number of cells in the region. Model classes include low, medium, and high site or survey potential zones as well as water, steep slopes, and mines. The percentages are valid both as percentages of total cells and percentages of land area (since each cell is the same size).
The number and percentage of random points occurring in each of the model classes. The percentages of random points in each class should be roughly the same as the percentage of grid cells in the same class.
The number and percentage of negative survey points occurring in each of the model classes. If surveys have been unbiased, they should exhibit a random distribution within the model classes. This would be evidenced as percentages in each class that are roughly the same as the percentages of random points and grid cells in the same class.
For the site probability models only, the number and percentage of modeled sites occurring in each of the model classes. The more their distribution (percent of sites in each class) deviates from random, the better the model has been able to detect their pattern.
For the site probability models only, the number and percentage of test population sites (Section 7.5.1.3.3) occurring in each of the model classes. Their pattern provides the best evaluation of how well the model performs.
For the survey probability models only, the number and percentage of all sites (at the time of modeling, including single artifacts) occurring in each of the model classes.
For the survey probability models only, the number and percentage of all known sites (at the time of modeling, including single artifacts) and all negative survey points occurring in each of the modeled classes. This is the population of surveyed places used to build the model.
The gain statistics (Section 7.5.1.2.3) for the training and test populations are calculated from the values in this table.

The third table reports the conditional Kappa values for each site or survey potential class (Bonham-Carter 1994: 247). This shows the agreement of the two preliminary models broken down by site or survey potential class. The conditional Kappa (K_I) is adjusted by the expected agreement due to chance alone. The proportion correct reported in the third column is a simple estimate of the correct proportion for each class and is not adjusted by the expected agreement. The table shows that the simple proportion correct overstates the amount of correlation between the two models.

Return to Top

8.4 MODEL INTERPRETATION

The model interpretation approach, procedures, and other considerations are discussed in Section 7.6. This section explains how the results of the interpretations are presented in Section 8.6.3 and Sections 8.7 – 8.26, and provides a discussion of variables previously identified as being associated with archaeological site locations in Minnesota.

8.4.1 Presentation of Interpretations

Various statistical analyses were performed to elicit differences between sites or surveyed places and random points, to evaluate the relationship between site potential and individual variables, to determine the independence of variable pairs, and to examine differences between groups of sites with different archaeological characteristics. These are discussed for each modeling region's site and survey probability models. The main focus of the interpretation was placed on the site probability models because of time constraints.

Model interpretations are supported by:

A table of model variable statistics providing:
- Spearman's rank order correlation coefficients (Section 7.6.2) of relationships between individual variables vs. sites and surveyed places.
- Significance values from the Mann-Whitney U test (Section 7.6.2) for each variable vs. modeled sites and surveyed places. These values supplement the comparison of means between sites and random points in the sense that there is generally very little difference between means if the Mann-Whitney U test is not significant.
- Mean and standard deviation of each variable for the populations of modeled sites, sites in low site potential areas, negative survey points, and random points.
A set of histograms graphing the distribution of the values of individual model variables for random points, negative survey points, and modeled sites. Horizontal distances and size of lake or watershed values are reported as square roots. Square these values to determine true distance and size values.
A variable correlation matrix showing Spearman's rank order correlation coefficients (Section 7.6.2) for all possible variable pairs. These are calculated from variable values for the population of modeled sites only. They do not necessarily indicate the overall correlation of variables throughout the region as a whole. Correlations at or close to zero indicate the variables are independent of one another.
A table showing the distribution of site potential for several dichotomous archaeological variables and results of chi-square tests of the significance of the relationship between site potential and these variables. P values of 0.05 or less are considered significant.

There are several things to keep in mind when considering these statistics:

Positive correlations (those where the coefficient is a positive number) indicate that the larger the value of the variable, the higher the site potential is likely to be. In other words, there is a direct relationship. If distance variables have positive coefficients, then sites are likely to be found far from the feature (i.e. sites may be farther away from pine forests than are nonsites).
Negative correlations (those with negative coefficients) indicate that the smaller the value of the variable, the higher the site potential is likely to be. This is referred to as an inverse relationship. If distance variables have negative coefficients, sites are likely to be near the feature (i.e. sites may be nearer to lakes than nonsites).
All horizontal distance variables are expressed as square roots of meters. Square these numbers to get the actual distances.
All elevation, vertical distances, and height variables are in feet.

These analyses are based on one model variable at a time, whereas the models are multivariate in nature. Consequently, these statistics, especially those relating to variable correlations, are to be interpreted cautiously.

8.4.2 Previously Identified Variables

Over the past 20 years, a number of researchers identified specific variables and site types used to create successful predictive site models. Many of these variables play a role in the models developed for Minnesota. Rather than reiterate the results of these previous modeling efforts along with the interpretation of each model, they are discussed in this section to provide a context for interpretation of the Mn/Model results. The variables are grouped according to six major categories: hydrologic, topographic, vegetation, soils, landforms, and seasonality. Although a review of the variables can proceed according to these classes, the results of numerous modeling studies indicate that many variables form an interrelated set of predictors that cross-cut any simplified classification scheme.

8.4.2.1 Hydrologic

The single most important set of variables for predicting the location of precontact archaeological sites is proximity to water resources. Lake basins and stream valleys frequently possess greater topographic relief and vegetation diversity than their surroundings, adding to the importance of these areas as locational predictors. A somewhat different set of water resources is involved at different periods, including glacial/post glacial lake strandlines (Paleoindian), rivers, large lakeshores/marshes and third-order streams (Archaic), and rivers and lakes (Woodland and Iroquois).

Cassell et al. (1997:101) include distance to water and riverine setting as two of three variables used to predict sites in Camp Ripley, Minnesota. Six hydrologic variables (lakeshore/beach ridge, within 50 meters of a river or stream, within 150 meters of a lake or lakeshore, within 150 meters of a known or suspected wild ricing or fishing area, immediately adjacent to a wetland, and rise of dry land adjacent to or within a marsh where plant resources occur) along with one landform variable form the basis of an intuitive model used on a number of pipeline projects in Minnesota (Dobbs and Mooers 1990:11-12; Dobbs et al. 1994:16-17).

Dalla Bona (1994:28-29) and Dalla Bona and Larcombe (1996) use stream order, distance to rapids and falls, lakes, and attributes of lakes (size, depth, and type of bottom) in the construction of an intuitive predictive model for western Ontario based on ethnographic information. Proximity to a variety of water features characterizes many site locations in southern Ontario (Young, et al. 1995:26-36). A model developed for the upper Potomac River in West Virginia indicates that distance to nearest stream (together with slope) is strongly correlated with site location (Neumann 1992:117). In two studies of site locations in Illinois, Warren (1990:209) and Warren and Asch (1996:9) find that high site probability areas occur on elevated creek terraces and along the valleys of headwater streams. Limp et al. (1981:75-78) and Lafferty, et al. (1985:207-210), in studies of site location in southeastern Arkansas, conclude that they are nearer water than would be expected by chance. Sites in the Sierra Nevada foothills of California tend to be near water in 78 percent of the cases in an analysis by Pilgrim (1987). In this study, 35 percent are near perennial streams, 25 percent near springs, and 18 percent near intermittent steams. Sites in the Reese River Valley of Nevada occur within 1000 meters of semipermanent water (Williams et al. 1973:227). Shermer and Tiffany (1985) explore the relationship between site locations within the Iowa River valley and a number of variables. They conclude that sites are disproportionately located within 500 meters of water.

A point system is the basis for an intuitive model used in New Jersey (Hasentrab 1991). In this model, site attributes, such as proximity to a tributary-river confluence, a river-river confluence, a river, a tributary, or a wetland, are given numerical scores. The scores are added, resulting in a maximum site sensitivity value of 130. The Vermont State Historic Preservation Office uses a more complex system. It includes 26 variables, nine of which pertain to surface hydrology. Any given location is assigned a point rating based on its proximity to various hydrologic features, such as confluence of intermittent streams, or falls/rapids. Distances are usually divided into three groups (0-60 meters, 60-120 meters, and 120-180 meters). The points are added, with each area given a designation of "sensitive" or "not sensitive".

Kvamme (1992) employs horizontal and vertical distances to water or entrenched drainages in his predictive model of sites on the high plains of Colorado. Finally, a predictive model developed for the area around the Seim/Livingood site (21CP29) near Granite Falls, Minnesota concludes that there is an increased chance of finding sites with lesser horizontal distances from water, and lesser horizontal and vertical distances from the Minnesota River (Larson et al. 1991:10.10).

8.4.2.2 Topographic

Nearly all of the topographic variables used in predictive site modeling are based on digital elevation models (DEM’s). In most predictive models, topographic features tend to be highly correlated with those related to surface hydrology. For example, since rivers, streams and lakes are usually set within their own valleys or basins, they have greater relief compared to their immediate surroundings. It is not surprising that many of the studies cited above also identify various topographic variables as being important in site prediction.

The amount of relief plays a role in many models. A number of models find that sites are limited to relatively level ground, usually with slopes of 10 - 15 percent or less (Cassell et al. 1997:101; Dalla Bona and Larcombe 1996: Figures 12-4, 12-6, 12-8; Kvamme 1985:223; 1992:29; Neumann 1992:117; Pilgrim 1987:36, 39; Williams et al. 1973:15). Kvamme (1985:223-224), in a study of site locations in west-central Colorado, finds that sites tended to be in areas of less relief compared to nonsites. Other models predict sites in areas where there is moderate or high relief, closely corresponding to the edges along river or stream valleys and lake basins (Carmichael 1990; Kvamme 1992:29; Larson et al. 1991:10.10; Warren and Asch 1996:9).

Aspect, or the direction of slope at or near a site, is included in many models. A southern aspect, which offers more warmth from the sun, is thought to be a desired location for many sites. Models where southern aspects are thought to be a factor in site location include western Ontario (Dalla Bona and Larcombe 1996: Figures 12-6, 12-8), Maine (Kellogg 1987:146), and southeastern Colorado (Kvamme 1985:223; 1992:29). The analysis of central Iowa by Shermer and Tiffany (1985:222-223) was unable to establish a link between southern exposures and site location.

Two variables that are closely related to local relief and aspect are shelter index and viewshed. Shelter from wind, inclement weather, or excessive sunlight are identified by Jochim (1976:49-51), Kvamme (1991:133) and others as a key element in site location. The Colorado studies by Kvamme (1985:223; 1992:29) give mixed results: extended activity sites in the west-central part of the state are more protected than nonsites while those in the southeast are more exposed than nonsite locations. Dalla Bona and Lacrombe (1996: Figure 12.8) identify sheltered land at the base of southern-facing hills near suitable moose habitat as one location for winter sites in western Ontario. Viewshed, or the desire to inhabit areas having a relatively large view of the surrounding landscape for viewing game, defensibility, and other reasons, is thought to be an important factor in hunter-gatherer settlement systems (Dalla Bona and Larcombe 1996:Figure 12.8; Dobbs and Mooers 1990:11; Jochim 1976:49, 51, 55; Larralde and Chandler 1981:135-136). In southeastern Colorado, Kvamme (1985:Table 1) finds little evidence for this relationship. However, in west central Colorado, both extended and limited activity sites are located closer to vantage points compared to their nonsite counterparts (Kvamme 1985:226). Sites in Chippewa County, Minnesota tend to have greater view angles than their nonsite counterparts (Larson et al. 1991:10.10).

8.4.2.3 Vegetation

Proximity to plant resources has long been thought to be an important factor in site location. Plants can be divided into edible and non-edible resources, the latter applying to wood used as fuel and raw materials (see Jochim 1976:51, 54-55; Lafferty et al. 1985:87-88). In the Upper Midwest, Dobbs and Mooers (1990: 11-12) maintain that archaeologically-sensitive areas are within 150 meters of known or suspected wild ricing areas or near areas thought to produce cranberries or blueberries. However, there is no conclusive relationship between the location of sites and nut resources in Arkansas (Limp et al. 1981:76-77).

Other researchers indicate that sites tend to be located near a diversity of resources typical of ecotones (Brandt 1992:275; Lafferty et al. 1985:71; Shermer and Tiffany 1985:222). Many of these resources are situated along stream valleys and lake margins where topographic relief, shelter, and levels of aquatic resources are also high. This is particularly true of semiarid environments where riverine ecotones are characteristically high in plant and animal diversity. However, there is no relationship between site location and proximity to wood resources or ecotones in west central Colorado (Kvamme 1985:224) and central Iowa (Shermer and Tiffany 1985:224, 226).

8.4.2.4 Soils

The variable of well-drained soils is incorporated into several predictive site models. Two of these are intuitive models that have yet to test this relationship (Dalla Bona and Larcombe 1996: Figure 12.6; Hasentrab 1991:51). Two other studies by Warren and Asch (1996:9) and Shermer and Tiffany (1985:223) from the upper Midwest conclude that sites tend to be located on well-drained soils.

8.4.2.5 Landforms

Several predictive models identify particular landforms associated with the location of archaeological sites. Craig (1989) finds that there is a shift from locations on sandridge terraces during the Archaic, to sandridge terraces/river bottoms during the Woodland period, and finally to river bottoms in Mississippian times. Howes (1982) maintains that large summer or winter campsites in the Alberta foothills occur on low river terraces while summer sites are more evenly spread throughout the region. The probability of finding sites in the prairie of Illinois is greatest on upland knolls and floodplains, among other areas (Warren and Asch 1996:9). The Vermont SHPO also incorporates landforms, such as knoll/crest/promontory or kame/outwash terraces, into their predictive model.

8.4.2.6 Seasonality

A key consideration in nearly all of the models reviewed here is the distribution of plant and animal resources throughout the year. Since most predictive models are based on variables related to the physical and biotic environments, it is clear that seasonal variations in the availability and location of the resources that people rely on for survival also have an effect on where those people live throughout the year. Although archaeologists discuss this topic (Kohler 1988; Neumann 1992:117; Young 1995:28; Williams et al. 1973:234), there are only a few models that directly incorporate seasonality (Dalla Bona and Larcombe 1996; Jochim 1976). This is because it is very difficult or impossible to determine when sites were occupied. Incorporating seasonality should result in better performing and more interpretable models.

Return to Top

8.5 MODEL COMPARISON

This interpretation of the model results is approached on several levels. The first is to compare the variables used for model construction from one subsection to another. Because of the large number of subsections and variables involved, this task is cumbersome. A reduction in the number of possible comparisons is accomplished by combining subsections. A second approach is to conduct a brief catchment analysis using average distances of variables from sites. A third analysis summarizes the locations of various site types in an effort to determine if there are significant differences between them. Finally, the basis of many intuitive reconnaissance surveys is examined by determining if sites are located within a certain distance of water. These approaches are undertaken in the following analysis.

8.5.1 Subsection Group Approach

The archaeological summaries of the 20 individual subsections or combined subsections can be interpreted by comparing them on the basis of a number of select variables. One way to begin is to compile the number of variables in each model, organized into four classes (hydrologic, topographic, vegetation, and soils). Table 8.5.1 presents the results of this analysis, organizing the subsections by the number of hydrologic variables involved in each model. This variable class is used because it plays a significant role in nearly all models. It is also the single most important class of variables in most intuitive or expert archaeological models. Consideration of intuitive models is necessary since they form the basis for where archaeological investigations (and the discovery of sites) have been conducted.

Table 8.5.1. List of Model Subsections Arranged by the Number of Hydrologic Variables.

Subsections	Hydrologic Variables	Topographic Variables	Vegetation Variables	Soils Variables
Mille Lacs Uplands	9	4	5	1
Border Lakes	8	4	4	0
Aspen Parklands	8	4	6	2
Pine Moraines and Outwash Plains	8	2	3	0
Chippewa Plains	7	3	4	0
Hardwood Hills	7	2	4	2
Big Woods	7	3	4	2
Minnesota River Prairie	7	2	6	2
Coteau Moraines/Inner Coteau	7	3	1	0
Anoka Sand Plain	6	2	3	1
L. Superior/North Shore/Nashwauk Uplands	6	1	7	0
St. Louis Moraines	5	3	2	1
Oak Savanna	5	2	4	0
Red River Prairie	4	4	4	3
Agassiz Lowlands	4	4	8	2
Littlefork-Vermilion Uplands	4	2	1	1
Blufflands	2	6	5	1
Rochester Plateau	1	6	6	1
St. Croix Moraines and Outwash Plains	1	3	0	1
Laurentian Highlands	1	3	1	1

The order of subsections in Table 8.5.1 is based on the number of hydrologic variables in the models. Distinct geographic clusters of subsections are apparent along this continuum. These divisions are indicated in the table. Four subsections (Mille Lacs Uplands, Border Lakes, Oak Savanna, and Laurentian Highlands) do not follow this pattern. In all cases, these four subsections were combined with others nearest to them geographically and most similar in composition of model variables (Figure 8.2). Mille Lacs Uplands was combined with the Aspen Parklands, Pine Moraines and Outwash Plains, Chippewa Plains, and Hardwood Hills subsections to form a "north-central" group. Border Lakes and Laurentian Highlands were consolidated with St. Louis Moraines and Glacial Lake Superior Plain/North Shore/Nashwauk Uplands, forming a "northeast" group. Finally, Oak Savanna and Minnesota River Prairie were combined with Big Woods, Coteau Moraines/Inner Coteau, and Anoka Sand Plain subsections to form a "south" group. Two additional groups were defined for the "northwest" (Red River Prairie, Agassiz Lowlands, Littlefork-Vermilion Uplands) and "southeast" (Blufflands, Rochester Plateau, St. Croix Moraines and Outwash Plains) portions of Minnesota.

It is apparent from this organization of subsections that when hydrology is the only class under consideration, the model results group the subsections in approximate geographic order. Where there is a low diversity of hydrologic features (Blufflands, Rochester Plateau), topographic and vegetation variables play a more important part in model development. There is no well-defined geographic clustering of subsections if they are arranged according to the number of topographic, vegetation, or soils variables. The groups of subsections derived from Table 8.5.1 are used in this section as an organizational device to characterize the models and also make comparisons across groups.

Table 8.5.2 lists 43 variables involved in 20 subsection models. The variables are grouped according to seven classes (lakes and wetlands, rivers and streams, undifferentiated water, topography, alluvial/lacustrine sediments, vegetation, and soils), while the subsections are assigned to five geographic classes (south, north-central, northwest, northeast, southeast). Variables that occur in half or more of the models for any group of subsections are in shaded in yellow.

Table 8.5.2. Model Variables Arranged According to Subsection and Subsection Groups (X = present in model, blank – not present).*

Variable

Class/

Variable

South

North-Central

Northwest

Northeast

Southeast

Coteau Moraines /Inner Coteau

River

Prairie

Big

Woods

Anoka

Sand

Plain

Oak

Savanna

Aspen

Parklands

Chippewa

Plains

Pine

Moraines

and

Outwash

Plains

Hard

-wood

Hills

Mille

Lacs

Uplands

Red

River

Prairie

Agassiz

Lowland

Littlefork-

Vermillion

Uplands

Laurentian

Highlands

St. Louis

Moraines/

Tamarack

Lowlands

Lake

Superior/

North

Shore/

Nashwauk

Uplands

Border

Lakes

St. Croix

Moraines

and

Outwash

Plains

Bluff-

lands

Rochester

Plateau

Lakes and

Wetlands

Distance to

nearest

permanent

lake I/O

Distance to

nearest

lake I/O

Distance to

nearest

permanent

wetland I/O

Distance to

nearest large

lake

Distance to

nearest

swamp

Distance to

nearest big

wetland

Size of

nearest lake

Size of

nearest

permanent

lake

Rivers and

Streams

Conf between perm/interm

streams &

lg rivers

Distance to

nearest

perennial river

or stream

Distance to

nearest

intermittent

stream

Undifferen-

tiated Water

Direction to

nearest water

or wetland

Vertical distance

to nearest water

Distance to lake/wetland

/organic soil/

stream

Vertical distance

to permanent

water

Alluvial/

Laucustrine Sediments

On holocene

alluvium

Distance to

glacial lake sediment

Topography

Slope

Absolute

elevation

Relative

elevation

within 90

meters

Height above surroundings (within 90 m)

Surface

roughness

within 90

meters

Prevailing orientation

or aspect

Distance to

minor ridge

or divide

Distance to

major ridge

or divide

Size of major watershed

Size of minor watershed

On river

terraces

Vegetation

Distance to

paper birch

Distance to

Big Woods

Distance to

sugar maple

Distance to conifers

Distance to

prairie

Distance to

aspen-birch

Distance to hardwoods

Distance to

oak

woodlands

Distance to

mixed

hardwoods

and pines

Distance to

brush lands

Distance to

river bottom

forest

Distance to

pine barrens

or flats

Vegetation diversity

within 990

meters

Soils

Distance to

well-drained

soils

Distance to

organic soil

*Yellow shading indicates variables that occur in half or more of the models for any group of subsections.

8.5.1.1 Southern Group of Subsections

This area includes most of the southern portion of the Minnesota and NE Iowa Morainal section of the Eastern Hardwood Forest province and all of the North Central Glaciated Plains section of the Prairie Parkland province. Nearly half of the variables in the models for these five subsections are related to the area's lakes, wetlands, rivers, and streams. Lakes and wetlands play the largest role in model development in southern Minnesota, with eight variables involved in the five models. Two of these, distance to nearest lake inlet/outlet and distance to nearest permanent lake inlet/outlet, are present in three of the five. Values are highly variable, ranging from an average distance of about 1.5 – 7.2 kilometers to lake inlet/outlets to a range of 8.8 – 38 kilometers to permanent lake inlets/outlets. This difference probably reflects the smaller number and lower density of permanent lake inlet/outlets in the landscape. These two variables appear in the other area models less frequently or not at all. Although the reasons for this are not entirely obvious, this area has an intermediate density of both lakes and streams when compared to other parts of Minnesota. This suggests that inlet/outlet variables may not be as significant where lakes are rare or where lake/stream intersections are so common as to be meaningless. Given this, it is not difficult to understand why distance to the nearest perennial river or stream, and distance to nearest large lake also play a role in model construction in the south. Mean distance to nearest perennial rivers and streams ranges from about 0.44 to 1.2 kilometers while distance to nearest large lake is more variable (1.3 – 10.6 kilometers), probably because large lakes are less common in the area's landscape. Distances to large lakes and to perennial rivers and streams are also involved in the majority of models from the north-central group of subsections. Vertical distance to water of any kind also plays a role in the construction of three models in the south, but not in any other area except the northeast (Border Lakes subsection). Sites in the south tend to occur on elevated areas averaging about 24 to 26 feet above water. Distance to any type of water (lakes, wetlands, streams) or organic soil is involved in four out of five of the southern subsections. The average distance ranges from approximately 50 to 115 meters. This variable also plays a role in a majority of north-central and northwest subsection models.

Three topographic variables are used to construct three of five models in the south. Height above surroundings within 90 meters is used in four southern models in addition to the majority of models from the north-central, northwest, and southeast areas. Sites in the south range from about 10 to 15 feet above their surroundings. Size of major watershed is also important in site location, ranging from 192 to 274 hectares (475 – 675 acres). The southeast area is the only other one that includes this variable in more than half of the models. A final topographic variable, on river terraces, is involved in three southern models. Values for this variable can be 1 (terrace present) or 0 (terrace not present). River terraces, at the scale mapped, occupy a greater areal extent of the southern area than any other part of Minnesota. In the south, 8.9 percent of modeled sites are on terraces, while the figure for the rest of the state is 5.6 percent. There are no other areas of the state this variable plays such an important role.

Three vegetation variables are involved in three models in the southern area. One of these, distance to sugar maple, is used in only one model for each of the four other areas of the state. Average distances range from approximately 1.25 to 14.4 kilometers. Distance to hardwoods has a much narrower range, from an average of about 60 meters in the Big Woods to 1.76 kilometers in the Minnesota River Prairie. As with the hydrologic variables discussed above, wider ranges of values are to be expected for features that are more scarce in the landscape than the features to which they are being compared. Once again, the southeast is the only other area of the state to have this variable involved in more than half of the models. Vegetation diversity within 990 meters plays a role in the Coteau Moraines/Inner Coteau, Minnesota River Prairie, and Oak Savanna subsections. This variable, which has possible values ranging from one to four for the number of vegetation types present, averages 1.8 in the Coteau Moraines/Inner Coteau to 2.5 in the Oak Savanna. Vegetation diversity is also used in a majority of models from the north-central and northwest parts of the state.

8.5.1.2 North-Central Group of Subsections

The north-central group includes the entire Lake Agassiz/Aspen Parklands section and the northern portion of the Minnesota and NE Iowa Morainal section of the Eastern Hardwood Forest province, as well as the western portion of the Northern Minnesota Drift and Lake Plains section and the entire Western Superior Uplands section of the Laurentian Mixed Forest Province. There are six hydrologic variables involved in three or more models from the north-central area. Four of these are involved in all subsection models. The mean distance to nearest large lake ranges from 0.53 to 2.02 kilometers for four of the subsections; the Aspen Parklands, which has few lakes, has a significantly larger value of 32.8 kilometers. Except for this latter subsection, distances to the nearest large lake in the north-central area are generally lower than they are in the south or northeast, reflecting the relatively high number of large lakes present in the area. This variable is also used to construct a majority of models in the south and northeast areas of the state. Distance to nearest large wetland is involved in three north-central models but does not play as significant a role in the other areas. Average distances vary from 0.26 to 2.7 kilometers. The average distance to nearest perennial river or stream is tightly constrained for the five models (0.73 – 1.22 kilometers). These figures are close to those of two other areas (south, northwest) where the variable plays a role in a majority of models. The sine of the direction to nearest water or wetland plays a part in the construction of all five subsection models. Mean values range from 0.11 to 0.34 which means that sites tend to be either east or west of water in a slightly southerly direction. This variable is also involved in three out of four models in the northeast area. Vertical distance to permanent water is used to construct all five models. Except for the average of 18.3 feet for the Hardwood Hills subsection, all values form a tight cluster between 8.5 and 10.8 feet. Only one other area, the northwest, has this variable in a majority of its models. Average distance to the composite variable of lake/wetland/organic soil/stream varies from a low of 35 meters to a high of 81 meters. This variable is also used to construct most if not all models from the northwest and southern areas of Minnesota.

Three topographic variables are involved in the construction of a majority of subsection models. Absolute elevation of sites ranges in mean value from 1047 to 1321 feet, while the individual subsection elevation means range from 1086 to 1400. Thus, sites tend to cluster around or slightly below the mean elevation, which tends to be higher than in the northwest and southeast areas where the variable is also important in site location. Height above surroundings within 90 meters is used in all five north-central models and a majority of those from all areas except the northeast. The mean values, ranging from 4.5 to 14.7 feet, are comparable to other areas. The average distance to minor ridge or divide ranges from about 465 to 565 meters for three subsections. The southeast part of the state is the only other area of the state where a majority of models includes this variable.

There are two vegetation variables, distance to prairie and vegetation diversity, used in the construction of a majority of models from the north-central area of the state. Average vegetation diversity for three models ranges from 2.3 to 3.2, similar to figures in the south and northwest areas. All five subsection models have distance to prairie as one of their variables. The mean distance is highly variable, ranging from about 425 meters in the Aspen Parklands, where prairies are quite extensive, to 85.3 kilometers in the Chippewa Plains, where prairies are few and small. There are no other areas where a majority of the models include this variable.

8.5.1.3 Northwest Group of Subsections

This group of three subsections lies within the former lakebed of Glacial Lake Agassiz, though the Red River Valley section is in the Prairie Parkland province while the other two subsections are in the North Minnesota and Ontario Peatlands section of the Laurentian Mixed Forest province. The numbers of variables involved in at least two of the models are evenly divided into hydrology, topography, and vegetation. Values for distance to nearest permanent wetland inlet/outlet are very high, ranging from 21.9 to 33.9 kilometers, perhaps indicating an avoidance of very wet places. This variable is not involved in a majority of models from the other areas. The average distance of sites to the nearest perennial river or stream is significantly closer at 730 to 925 meters. This figure is comparable to the ranges of other subsection groups. Distance to any lake, wetland, organic soil, or stream is much less, averaging between about 25 and 95 meters. Again, these figures are similar to other subsections. Vertical distance to permanent water is between 6.6 and 9.2 feet, at the lower end of subsections in the north-central group. However, this area is mostly lake plain with very low relief. The average values of 0.20 to 0.34 for the nominal variable on Holocene alluvium are consistent with the 29.43 percent of sites in the northwest group being situated on alluvium, which makes up only 2 percent of the landscape. There are no other subsection groups that have this variable participating in a majority of the models.

Three topographic variables are used to construct models in a majority of the northwest models. Average elevation at archaeological sites ranges in value from 950 to 1185 feet, while mean elevations range from 754 to 10075. Thus sites tend to be found at higher than average elevations in the area. The site elevation means are comparable to subsection averages in the north-central and southeast parts of the state. Height above surroundings within 90 meters is 6.4 and 10.1 feet, also similar in average value to other subsections. Finally, average surface roughness ranges in value between 150 and 180. These values are in the lower range of surface roughness in the area as a whole, where roughness ranges from 117 to 441. However, they are near or slightly lower than the mean values for the area as a whole. In the southeast part of the state, these figures are about the same for two of its subsections.

There are also three vegetation variables involved in two or more of the northwestern subsections. Two of these, distance to aspen-birch and distance to hardwoods, are not involved in a majority of subsection models from other areas of the state. The mean distance of archaeological sites to aspen-birch is highly variable, ranging from 440 meters in the Agassiz Lowlands to 61.5 kilometers in the Red River Prairie, where few, very small stands of aspen-birch are mapped only along the eastern edge of the subsection. There is less variation in distance to oak woodlands, from 32.0 kilometers in the Red River Prairie to 71.8 kilometers in the Agassiz Lowlands. The values of vegetation diversity (2.7 – 2.9) in the northwestern area fall between those of the southwest and north-central group of subsections.

Unlike other subsection groups, distance to organic soils plays a role in the development of a majority of models in the northwest. Once again, the values are highly variable at 1.5 kilometers in the Agassiz Lowlands to 27.6 kilometers in the Red River Prairie. This points to the fact that organic soils are common and widespread in the Agassiz Lowlands.

8.5.1.4 Northeast Group of Subsections

These northeastern subsections include the entire Northern and Southern Superior Uplands sections and the eastern portion of the Northern Minnesota Drift and Lake Plains section. All are within the Laurentian Mixed Forest province. This area of the state is unique for there are only three variables (all hydrologic) that are involved in a majority of the models. The average distance from the nearest large lake archaeological sites are located ranges from 121 to 576 meters, at the low end of values for other subsections. This may reflect the number of large lakes in the area or simply that large lakes have been the focus of surveys here. Distance to nearest swamp plays a role in all four subsection models but is not involved in a majority of models from other areas of the state. Average site distances range from a low of 169 meters in the Glacial Lake Superior/North Shore/Nashwauk Uplands to a high of 484 meters in the Laurentian Highlands. As swamps occupy approximately 75 percent of the landscape in this part of the state, these values indicate a tendency for sites to be farther from swamps than the average piece of real estate. Direction to nearest water or wetland plays a role in three subsection models. Mean values range from 0.18 to 0.28, indicating sites are located east or west of lakes in a slightly southerly direction, as in the north central subsections discussed above..

8.5.1.5 Southeast Group of Subsections

The southeast group of subsections are entirely within the Eastern Broadleaf Forest province, including all of the Paleozoic Plateau and a small part of the Minnesota and Northeast Iowa Morainal sections. In contrast to the northeast part of the state, hydrologic variables do not play a role in the majority of the models from southeast Minnesota. Rather, variables related to the area's topography and vegetation do. The first of these is elevation, which is in two models and averages from 830 to 1193 feet for sites, similar to the landscape as a whole, which averages 912 to 1180 feet. These averages are also within the range of variation of other subsection models. Height above surroundings within 90 meters is involved in all southeast models. The mean values reach a low of 12.6 feet in the Rochester Plateau to highs of 22 and 23 feet in the other two subsections. This variable is also involved in a majority of models in the south, north-central, and northwest areas of the state. The figures for the southeast generally tend to be higher than in other subsections, reflecting the greater topographic relief in the area. Surface roughness has mean values of 149 and 150, comparable to the Red River Prairie and lower than the roughness of the terrain in the region as a whole. The average distance to minor ridge or divide ranges from 529 to 655 meters, at the high end for the north-central models. Distance to major ridge or divide is higher at 4.4 – 7.1 kilometers, though these values for sites hover around the mean distance for all cells in the area of 6.0 kilometers. There are no other areas of the state where this variable plays a role in the construction of a majority of models. Size of major watershed is involved in the Blufflands and Rochester Plateau models. Averages range from 278 to 380 hectares (688 to 940 acres), indicating that sites tend to be in larger than average watersheds for the area. These figures are also somewhat higher than the models from southern Minnesota.

Three vegetation variables are also used to construct at least two of the three models in the southeast area of the state. Average site distance to Big Woods ranges from 0.53 to 3.6 kilometers, while all cells in these sections average 1.6 to 5.0 kilometers from Big Woods. This variable plays a role in only two other subsection models. Distance to conifers is significantly higher, ranging from 138 to 161 kilometers. Since there is only one small patch of coniferous vegetation mapped in this area, near its northern boundary, this variable may be acting as a surrogate for the north-south axis and reflect the concentration of known sites in the southern part of the area. There is no other area that uses this variable in a majority of its models. Average distance to hardwoods is 169 meters in the Blufflands and 2.4 kilometers in the Rochester Plateau, much closer than the averages of 959 meters and 3.5 kilometers for all cells in the same subsections. These values fall in the range of those from the southern area of Minnesota.

Distance to glacial lake sediment plays a role in the Blufflands and Rochester Plateau models. Like coniferous trees, these sediments occur only near the northern boundary of the area. The mean site distance to these sediments (111 to 118 kilometers) again reflects the great preponderance of sites in the southeastern portion of the area.

8.5.1.6 Synopsis

The location of precontact archaeological sites in Minnesota is constrained by a small number of variables that span a variety of areas within the state. The most important of these relate to surface hydrology. There are a total of 106 variable occurrences in the 20 models. Dividing this by 300 (15 hydrologic variables x 20 models) yields a figure of 35.3 percent participation in the models. There are three variables that seem to be the most important in site selection over the largest number of models and broadest geographic distribution. The general category of distance to nearest lake, wetland, organic soil, or stream is present in 14 models. The average distance of archaeological sites to these water sources in models where this variable plays a role is 109 meters. Distance to relatively large and/or permanent hydrological features also plays an important role in site location. Distance to nearest large lake participates in 13 of 20 models. The mean distance to this hydrologic feature is 3.3 kilometers, substantially higher than the composite hydrology distance measure of 109 meters. Distances to perennial rivers or streams ranges from a low of 121 meters to a high of 32.8 kilometers. Twelve models incorporate distance to nearest perennial river or stream. The average distance of archaeological sites from these features is 783 meters. The variation in subsection models falls within a narrow range of 0.29 to 1.22 kilometers. These are small figures compared to distances to large lakes because of the more pervasive nature of perennial river and streams compared to large lakes.

There are also two topographic variables that play roles in many subsection models, absolute elevation and height above surroundings. These variables are closely linked, since archaeological sites tend to be about 50 feet lower in elevation than the overall environment (i.e. random points). This reflects the general location of sites near water sources, the landscape around which tends to be lower in elevation than the overall landscape. Within these settings, sites tend to be on slightly elevated ground, ranging from lows of 2.6 and 4.5 feet in northern Minnesota to 22 to 23 feet in the southeast part of the state. An average height above surroundings of about 12 feet seems to be the norm in the central and southern portions of the state. There appears to be a desire on the part of the Native American inhabitants of Minnesota to live on elevated areas close to water sources. Because of the broken, dissected, and irregular topography in these areas, vegetation diversity tends to be higher than the general surroundings. On a scale from one to four, vegetation diversity in the vicinity of archaeological sites is higher, at 2.7, compared to 2.1 for the overall environment (i.e. random points).

Combining all of the locational information summarized above, archaeological sites in Minnesota are situated within several hundred meters from the edge of all water sources, current and ancient. This includes lakes, wetlands, streams, and organic soils. Within these parameters, there is a preference to be within about three to four kilometers of large lakes and one kilometer of perennial rivers or streams. If sites are not located on these relatively permanent water sources, they are certainly nearby. Clearly, precontact activity tends to be centered around these hydrological features. In some areas of the state, proximity to lake inlet/outlets, swamps, and wetlands play a secondary role in precontact settlement patterns. These locations also happen to be lower in elevation than the general landscape. Within this setting, however, there is a clear preference for occupation in areas somewhat above their immediate surroundings. This reflects occupations along the edges of lake basins and stream terraces, but may also include low rises in an otherwise relatively flat landscape. This is also reflected in surface roughness, which tends to be more variable in the vicinity of sites compared to the overall, regional environment. Relatively high relief also characterizes the edges of water sources. This, in addition to the protection water provides from prairie fires, results in relatively high vegetation diversity. Other variables that are factors in site location on a more limited scale are proximity to river or stream alluvium and organic or well-drained soils. In summary, most of these variables form a complex of interrelated and partially overlapping or redundant elements of water, vegetation, and topography. All three sets of variables combine to make these features preferred for past human habitation.

8.5.2 Site Catchment Analysis

The foregoing summary indicates a wide range in mean values for many of the model variables. The fact that the same variable may have large values in one area and small values in another indicates that not all had an equal role in where precontact peoples lived and that distribution of resources within the landscape varies considerably across the state. Distance to various resources is the cornerstone of site catchment studies, a popular approach to site location in the 1970s and early 1980s. Basically, a catchment is the area around a site where the inhabitants derive their resources (Roper 1979:120). These areas are commonly depicted as a series of concentric circles around a site, representing the hypothesized limits to where its inhabitants can travel on foot within a specified period of time. The areas of certain environmental zones (e.g. floodplain, oak-hickory forest, lake) within these rings are often depicted for each site, providing a way of characterizing its environmental setting. The premise of site catchments (and predictive models like Mn/Model) is that people will choose to live in those areas where the energy expended to procure life-sustaining resources is minimized. The farther a resource, the more energy (in terms of travel time) is required to exploit it. People are willing to expend more energy procuring some resources than others (Roper 1979:121). Some, such as water, are so basic that distance is minimized. The shape, size, and location of catchment zones vary depending on the spacing, zonation, and seasonal distribution of resources (Roper 1979:121). The correlation between site location and various characteristics of the biophysical environment is the premise behind both catchment studies and predictive modeling.

A number of studies discuss the distances people travel to exploit certain resources. Lee (1969:61) states that the !Kung bushmen travel no more than about 10 kilometers in a day from their camp without packing drinking water and other supplies for an overnight stay. Activities in the arid climate in southern Africa are closely tied to waterholes. Flannery’s (1976) study of catchment areas in Mexico delimits three zones: (1) 0 - 2.5/5.0 kilometers for agriculture; (2) up to five kilometers for exploiting minerals and wild plants; (3) up to 15 kilometers for firewood, deer meat, and construction wood. Zvelebil (1983:88) defines four limits for various subsistence activities in the boreal zone of Europe (three kilometers for farming, five kilometers for fishing, 10 kilometers for hunting, 15 kilometers for swidden horticulture).

Sometimes, the total environment is divided into a number of zones depending on the amount of use. For example, Higgs (1975:ix) differentiates between territory (area immediately accessible and habitually used by the inhabitants of a site) and catchment (area from which the contents of a site are derived). Kvamme (1985) goes beyond the traditional predictive site modeling approach by constructing two separate models of extended and limited activity sites for a 580,000 acre area in western Colorado. The purpose of the analysis is to refine location models based on only one site type. Kvamme (1985:Figure 9.5) divides the total space into four nested or hierarchical subsets. From largest to smallest these are the total environment, accessible space, activity space, and settlement space. Settlement space encompasses the smallest area. Only those activities that require extended stays or multiple tasks are included here. Site locations areas might be determined by idiosyncratic or very specific characteristics such as view quality, shelter quality, and proximity to water and fuel. Extended activity sites characterized by relatively large quantities and varieties of artifacts are most often found in settlement space. Activity space includes a somewhat larger area, defined by the presence of limited or short-term task-specific activities such as animal butchering, game viewing, flint knapping, and game viewing. Resources in this area are immobile, or relatively so, including plants, small animals, and some fish. Sites usually found in these areas are small in area and contain few tool or artifact types, synonymous with Kvamme’s (1985:214-215) limited activity sites. Accessible space encompasses all areas that humans can reach. This includes most areas of Minnesota except very steep slopes. Watercraft provides access to lakes and rivers. In terms of resources, large animals and fish would be found in this zone, in addition to the resources in the smaller zones. Wide-ranging Paleoindian groups would probably occupy large areas of accessible space. All sites, including single or isolated artifacts, would be included in the accessible space. Beyond the accessible space are areas not accessible to humans and therefore lacking sites. Based on a series of seven variables (vertical distance to nearest water, vantage distance, view angle, shelter quality, exposure, slope, local relief), Kvamme (1985:232-235) demonstrates that environmental conditions shift from "favorable" to "unfavorable" as you move from extended activity sites to limited activity sites, isolated artifacts, and nonsites. In summary, site selection is viewed as a process of defining a successively more narrow range of locations suitable for habitation (Kvamme 1985:229).

Although the type of analysis that Kvamme performed is not possible with Mn/Model, a preliminary exploration of site catchment is useful. The analysis focuses on all distance variables used in the subsection models. All other variables, comprising a minority of the total number, are excluded. Figures 8.3a and 8.3b plot most horizontal distance variables for the 20 subsection models along a square root meter scale using the same mean values that appear in Section 8.6. Due to space restrictions, variables that have mean square root values greater than 200 meters (40 kilometers or 25 miles) are not plotted. Plotting the square roots instead of the original, untransformed values compresses the variables that are far from the sites relative to those that are closer. These environmental factors are so distant from the site so as to preclude them from having an effect on site location. Two vertical distance measures, height above water and surroundings, are also included in the figures. These variables are measured in feet. Three arbitrary catchment zones (10 kilometers and beyond, 10 kilometers to 2.5 kilometers, within 2.5 kilometers) are included on the figures. The boundaries between them correspond to a round trip distance of 20 kilometers (12.5 miles) that can be walked in a day (see Lee 1969:61) and half of this value. Although any number of other zones can be substituted, the purpose of these figures is to illustrate the distances of environmental features from the sites and how they compare from one subsection to another. In some part of Minnesota, these figures may be different due to travel by watercraft. The following discussion briefly describes some of the variables involved in the models.

Figures 8.3a and 8.3b indicate that sites are situated within the 10 kilometer catchment zone (one day round trip) of most features used to derive distance variables. Over half of the features outside of this zone (28 of 44) are related to vegetation, including 19 features beyond the 25 mile limit illustrated in these figures. This means that site location, while related in some fashion to vegetation, is not directly based on many of the vegetation variables appearing in the models. Rather, they reflect the environment within which people lived or, in some cases, act as surrogates for other factors. The same might be said for large lakes and permanent lake or wetland inlets/outlets, which are the dominant hydrologic features in the outer zone. These features are best viewed as defining the inner limits of accessible space, but beyond activity space, to borrow terms from Kvamme (1985). There are no topographic features in this outer zone.

The zone between 2.5 and 10 kilometers, probably less than one days round trip, contains 38 features. About one-third (17) are vegetation features, a reduction in the relative frequency from the outer zone. There is an increase in hydrologic features, including stream and river confluences, over the outer zone. Lake inlets/outlets or permanent lake inlets/outlets are also included in this zone. The only topographic variable present is distance to major ridge or divide. These variables probably reflect site location strategies related to short-term task-specific activities (i.e. activity space) tied to the occupation of base camps. This is also probably the case for some of the variables that fall at the outer limits of the inner zone.

The inner-most catchment area is within 2.5 miles of the sites. More than half of the features (56 of 92) are hydrologic in nature; the remaining are divided between vegetation, soils, and topography. The dominant feature is "water", a generalization consisting of lakes, wetlands, organic soils, or streams. The average distance to water in all subsections where this variable plays a role is less than 100 meters. There is little doubt that the settlement space, defined as those areas where activity is intense and prolonged over a relatively long period of time, is closely tied to water. Distance from perennial streams or rivers plays a role in 12 models, nearly as many as the generalized water variable. Average site distance is somewhat farther at about one kilometer. Distance to swamps is involved in all northeast area models, averaging 169 to 484 meters. Still, in the context of the landscape, this indicates a preference to perform activities away from swamps, which are on average within 99 to 331 meters of cells in this landscape. Large lakes are usually within one kilometer of sites in most of the north-central models. Vertical distance to water is also important in site location in a majority of models. Site averages range from about five to 25 feet above water. Height above surroundings is involved in all except three models. Sites are usually 5-15 feet above their immediate surroundings. These and several other variables (e.g. distance to swamps, glacial lake sediments) define the environmental characteristics that have the strongest influence on where sites are located within settlement space. Clearly, precontact peoples preferred locations in horizontal and vertical proximity to water, somewhat above their immediate surroundings, near perennial streams, large lakes, but somewhat away from swamps. Other factors proved important in choosing the general environment in which to settle. All of these variables are best interpreted within the geographic context of the individual subsection. This is attempted in Sections 8.7 through 8.26. With the addition of refined topographic variables, such as vantage distance, view angle, shelter quality, and exposure (Kvamme 1985), these predictive models will become more precise and increase our knowledge of where and why precontact peoples settled in Minnesota.

8.5.3 Cultural Context and Site Location

Although the subsection predictive site models are based on the presence or absence of sites, the interpretations of some models presented in Sections 8.7 through 8.26 include an evaluation of the distribution of sites by cultural context. A summary of these relationships is presented in Table 8.5.3 for those subsections that contained a sufficient number of sites for a meaningful analysis. It is apparent from this table that sites assigned to the Archaic tradition and those without pottery (aceramic) are distributed in a significant non-random fashion in nine subsections. Specifically, Archaic and aceramic sites are over-represented in the low, or low and medium probability areas in seven of the twenty subsections compared to overall site distributions. Conversely, these sites are underrepresented in the high potential areas. There are several implications of these patterns. First, reconnaissance surveys focusing on areas of high and medium potential, at the exclusion of low areas, may be missing proportionately more sites with Archaic components than those of other time periods. Second, the tendency of sites in low potential zones to lack pottery may reflect the use of these areas for sporadic, short-term activities. Most of these sites are classified as lithic scatters and probably represent the habitation of the outer part of the activity space and parts beyond in the accessible space. Areas of high potential tend to contain higher proportions of sites with ceramics, pointing to a more settled, long term use in these locales. Most of these sites are situated in settlement space as defined above. In short, the probability areas defined by Mn/Model appear to tap into some underlying pattern of site settlement systems that can be related to past ethnographic work and site catchment studies.

Table 8.5.3. Summary of the Results of the Analysis of Cultural Context and Site Type Variables by Subsection.*

Subsection	Paleoindian	Archaic	Woodland	Plains Village	Oneota	Mississippian	Aceramic	Single Component	Mounds
Minnesota River Prairie	0	0	0	X	0	0	X	0	0
Hardwood Hills	-	0	X	-	-	-	0	0	0
Coteau Moraines – Inner Coteau	-	0	X	0	-	-	0	0	-
Pine Moraines and Outwash Plains	-	X	-	-	-	-	0	0	0
Red River Prairie	-	X	-	-	-	-	X	0	0
St. Louis Moraines – Tamarack Lowlands	-	X	-	-	-	-	X	0	-
Mille Lacs Uplands	0	0	-	-	-	-	X	0	0
Big Woods	-	X	-	-	-	-	X	X	0
Border Lakes	-	0	-	-	-	-	0	0	-
Anoka Sand Plain	-	0	-	-	-	-	X	0	-
Chippewa Plains	-	0	-	-	-	-	0	0	0
Blufflands	-	X	0	-	0	0	X	0	X
Oak Savanna	-	0	-	-	-	-	0	0	-
St. Croix Moraines and Outwash Plains	-	X	-	-	-	-	X	X	-

* Cell symbols designate a significant (X) and not significant (0) relationship based on chi-square. Cultural contexts with sample sizes too small to evaluate are designated by dash (-). Site sample sizes for six subsections (Rochester Plateau, Laurentian Highlands, Aspen Parklands, Littlefork – Vermilion Uplands, Glacial Lake Superior Plain – North Shore Highlands-Naswauk Uplands, Agassiz Lowlands) were too small to permit an examination of any of the cultural variables and are excluded from the table.

It is likely that the pattern in Minnesota is one repeated in many other parts of the world where base camps are located in the most desirable (high potential) areas, with short-term special activities tied to the base camps being performed in areas of medium to low potential. As a generalization there are exceptions, as demonstrated by the fact that there is no significant difference in the distribution of ceramic versus aceramic sites in a majority of subsections. Based on the same reasoning, there is an expectation that sites containing single components should be less likely to occur in high potential zones than multiple component sites. This is because sites located in areas of high site potential should have a relatively high chance of being reoccupied. This is clearly not the case, since the only two subsections with significant relationships (Big Woods, Pine Moraines and Outwash Plains) have a higher proportion of single component sites in the high potential areas compared to multiple component sites. The tendency for base camps to be in high probability zones could be one reason why sites identified by a Plains Village tradition occupation (i.e. Plains Village ceramics present) are underrepresented in the low and medium probability areas and over-represented in the high probability areas of the Minnesota River Prairie subsection.

8.5.4 SHPO Intuitive Model

The single most important variable used by Scott Anfinson of the Minnesota SHPO office in his intuitive model of precontact site location is distance to present and past water sources. The actual distance he uses in his review and compliance responsibilities has undergone some changes over the years. An early horizontal distance from water measure of 1,000 feet (305 meters) has recently been replaced by a figure of 500 feet (152.5 meters). In general, if an area is within 500 feet of the edge of water, the SHPO recommends that a Phase I reconnaissance survey is needed. It is instructive to use the data in the Mn/Model subsection databases to evaluate the usefulness of the 500 foot figure. Table 8.5.4 is a compilation of the composite hydrology variable "distance to nearest lake, wetland, organic soil, or stream" for the 20 subsections, ranked according to their median values. It also gives the means and standard deviations for this variable. Medians are used because they are the best measure of a variable that has a non-normal or skewed distribution. The distribution of original or untransformed values for all of the subsections is positively skewed (i.e. most values are low with a few outlying large distances). The order of subsections in Table 8.5.4 generally corresponds to location: sites in the watery landscapes of the north and north-central parts of the state tend to be nearer water than in the better drained central or southeastern portions. Except for the Laurentian Highlands, the median values are below 500 feet (152.5 meters), lending support to the SHPO figure. The mean values are higher than the medians due to positively skewed distributions. Table 8.5.4 also indicates that distance is highly dependent on location, with nearly half of the subsections below a median value of 30 meters (100 feet). The low figure for the St. Louis Moraines-Tamarack Lowlands and Border Lakes reflects the presence of many sites along the lakeshores in these subsections. Sites tend to be closer to water in the central part of the state compared to the south. Also, sites are about half the distance to water compared to randomly selected points. Generally, the more water bodies in a region the nearer sites are to water.

Table 8.5.4. Summary Statistics for Modeled Sites and Random Points by Subsection for the Variable Distance to Nearest Lake, Wetland, Organic Soil, or Stream (in meters).

Subsection	Modeled Sites				Random Points
Subsection	Median	Mean	S.D.	No. of Sites	Median	Mean	S.D.	No. of Sites
St. Louis Moraines – Tamarack Lowlands	0	35	78	186	67	212	380	3066
Border Lakes	0	102	256	960	60	115	200	2163
Littlefork – Vermilion Uplands	30	47	66	25	90	195	318	1288

Mille Lacs Uplands	30	70	114	437	60	213	744	3130
Pine Moraines and Outwash Plains	30	79	122	474	67	140	189	3978
Agassiz Lowlands	30	83	142	53	551	1507	2083	551
Chippewa Plains	30	95	161	513	85	255	538	2463

L. Superior-North Shore-Nashwauk Uplands	30	137	662	87	67	136	183	2651

Anoka Sand Plain	42	79	109	337	90	171	242	1141
Hardwood Hills	42	80	110	470	67	117	153	4294
Big Woods	60	94	108	637	85	134	162	1678

Aspen Parklands	60	131	171	59	180	344	544	2402

Coteau Moraines – Inner Coteau	85	123	128	350	162	245	234	1626
Red River Prairie	85	137	193	270	268	421	471	3894
Oak Savanna	90	138	194	121	242	343	351	1556
Rochester Plateau	90	165	172	81	170	197	167	1199
Blufflands	95	162	190	554	134	178	165	1200
Minnesota River Prairie	95	165	211	969	242	372	396	2404
St. Croix Moraines and Outwash Plains	114	192	228	126	127	251	323	800

Laurentian Highlands	211	319	288	120	131	106	67	396

State Total	60	116	192	6829	120	344	802	41880

Return to Top

Continue to 8.6 Model Results

The Mn/Model Final Report (Phases 1-3) is available on CD-ROM. Copies may be requested by visiting the contact page.

MnModel Orange Bar Logo

Acknowledgements

MnModel was financed with Transportation Enhancement and State Planning and Research funds from the Federal Highway Administration and a Minnesota Department of Transportation match.

Copyright Notice

The MnModel process and the predictive models it produced are copyrighted by the Minnesota Department of Transportation (MnDOT), 2000. They may not be used without MnDOT's consent.