Mn/Model Final Report Chapter 6: Archaeological and Environmental Variables

Quick Links

Chapters

Appendices

Chapter 6

Archaeological and Environmental Variables

By Elizabeth Hobbs and Tatiana Nawrocki

Statewide Survey Impelmentation Model Map

Chapter 6 Table of Contents
6.1 Introduction
6.2 Modern and Historic Environmental Variables
      6.2.1 Distance or Proximity Variables
      6.2.2 Terrain
      6.2.3 Surface Hydrology
      6.2.4 Soils
      6.2.5 Vegetation
      6.2.6 Historic Cultural Features
      6.2.7 Disturbance
6.3 Paleoenvironmental Variables
      6.3.1 Regular Periodic Character of Global Climate Change vs. Irregular Regional Variations
      6.3.2 Paleoclimate Model for Minnesota
      6.3.3 Palynological Data for Minnesota
      6.3.4 Geomorphology and Geology
6.4 Archaeological Variables
      6.4.1 Role of Lithic Scatters
      6.4.2 Site Absence
6.5 Conclusion
      References

6.1 INTRODUCTION

The premise of archaeological predictive modeling is that there is a quantifiable relationship between the presence of archaeological sites and environmental characteristics. If that relationship is statistically valid, then identifying similar environmental resources can be used to predict the probable locations of archaeological sites.

Environmental variables for modeling were derived, using spatial analytical techniques, from mapped layers of data. The data layers are descriptive, providing a representation of only one element of the landscape. For Mn/Model, these layers include elevation, surface hydrology, vegetation, soils, and geomorphology. The data layers selected for this project were the best digital information available.

The model variables are derived from interpretations of the data layers, making explicit the characteristics of the data layers that are relevant to archaeological site location. Variable definition is a critical aspect of model building. Although elevation itself may not be a good predictor of archaeological sites, variables derived from it (slope, height above surroundings, surface roughness) may be significant. Suggestions for environmental variables came from team members and literature reviews. The number of variables changed through the three stages of the project. The Phase 1 models were based on approximately 40 variables. The Phase 2 basic models drew from 69 variables. The Phase 3 models were based on 44 variables, as redundant or insignificant variables from Phase 2 were removed. Four new variables were introduced in Phase 3 to represent watershed geometry. Also in Phase 3, variables measuring distance to tree species mapped at a very low resolution were replaced by variables mapping distances to bearing trees recorded in the Public Land Survey.

Environmental data layers reflect current conditions or conditions in the recent past, all post-dating the sites in the archaeological database. As the environment is dynamic, these contemporary data may not be valid for past conditions. However, they act as surrogates for paleoenvironmental variables when information about the past is unavailable. This chapter will define and discuss the environmental variables that were used to construct the models (see Appendix B for more detailed information).

In the next two sections, modern and paleoenvironmental variables are discussed by category. A complete list of the environmental variables used for building the Phase 2 and 3 models is provided in Table 6.1.

Table 6.1: Environmental Variables Used in Phase 2 and 3 Statewide Modeling

Variables	Phase 2	Phase 3
*Variables derived from elevation*
Elevation	X	X
Height above surroundings	X	X
Prevailing orientation	X	X
Relative elevation	X	X
Slope	X	X
Solar insolation	X
Surface roughness	X	X
*Variables derived from watersheds*
Distance to nearest major ridge or divide		X
Distance to nearest minor ridge or divide		X
Size of major watershed		X
Size of minor watershed		X
*Variables derived from surface hydrology*
Direction to nearest water	X
Direction to nearest permanent water	X
Direction to nearest water or wetland	X	X
Distance to edge of nearest lake	X
Distance to edge of nearest large lake	X	X
Distance to edge of nearest permanent lake	X
Distance to edge of nearest river or stream	X
Distance to edge of nearest large river	X
Distance to edge of nearest perennial river or stream	X	X
Distance to nearest perennial stream	X
Distance to nearest intermittent stream	X	X
Distance to edge of nearest wetland	X
Distance to edge of nearest large wetland	X	X
Distance to edge of nearest marsh	X
Distance to edge of nearest swamp	X	X
Distance to nearest confluence between streams of different classes	X
Distance to nearest confluence between perennial or intermittent streams and large rivers	X	X
Distance to nearest confluence between perennial streams and large rivers	X
Distance to nearest lake inlet/outlet	X	X
Distance to nearest permanent lake inlet/outlet	X	X
Distance to wetland inlet/outlets	X
Distance to permanent wetland inlets/outlets	X	X
Distance to nearest lake or wetland inlet/outlet	X
Size of nearest lake	X	X
Size of nearest permanent lake	X	X
*Variables derived from elevation and surface hydrology*
Vertical distance to permanent water	X	X
Vertical distance to water	X	X
*Variables derived from low resolution soils*
Distance to edge of nearest area of organic soils	X	X
Distance to edge of nearest large area of organic soils	X
Distance to (from)* well-drained soils	X	X
Soil drainage	X
*Variables derived from surface hydrology and soils*
Distance to edge of nearest lake, wetland, or area of organic soils	X
Distance to (edge of)* nearest lake, wetland, (area of)* organic soil, or stream	X	X
*Variables derived from Marschner vegetation*
Distance to aspen-birch	X	X
Distance to Big Woods	X	X
Distance to brushlands	X	X
Distance to conifers	X	X
Distance to hardwoods (hardwood forest)*	X	X
Distance to mixed hardwoods and pines	X	X
Distance to oak woodland	X	X
Distance to pine barrens, (openings, and)* or flats	X	X
Distance to pine forest	X
Distance to prairie	X	X
Distance to river bottom forest	X	X
Distance to woods	X
Vegetation diversity within 0.5 km	X
Vegetation diversity within 1 km	X	X
*Variables derived from tree species distributions*
Distance to cranberry	X
Distance to Kentucky coffee tree	X
Distance to paper birch	X	X
Distance to sugar maple	X	X
*Variables based on disturbance factors*
Mine pits or dumps	X
Susceptibility to erosion by water	X
Susceptibility to sedimentation	X
*Variables derived from Quaternary geology*
Distance to glacial lake sediment	X	X
On alluvium	X	X
On colluvium	X
On lake sediment	X
On peat	X
On river terraces (On terraces)*	X	X
*Variables derived from bedrock geology*
Depth to bedrock	X
Distance to bedrock exposures	X
Distance to bedrock used for tools		X

* There were minor changes in names of several variables in Phase 3 as compared to Phase 2 models. Words in brackets refer to the Phase 2 names. Name changes in these cases did not impact content and procedures used to derive these variables.

Besides the variables used in statewide models and listed in Table 6.1, there were variables used in non-statewide models enhanced by soil and Trygg map data in Phase 2. These variables are described in Sections 6.2.4.3 and 6.2.5.4.

Return to Top

6.2 MODERN AND HISTORIC ENVIRONMENTAL VARIABLES

6.2.1 Distance or Proximity Variables

In general, the environmental data layers represent key resources for hunters and gatherers. These include lakes, streams, woodlands, potential wild rice sites, sugar maple, bedrock outcrops, and so on. Sites would not necessarily overlay these resources; e.g. a campsite would not necessarily have to be in a woodland, but would be expected to have had some source of fuel close at hand.

Of the resources mapped, some would not be places where settlements would be expected, for instance the interiors of lakes and wetlands. However, since water levels have fluctuated greatly over the last 10,000 years, some sites are now within lakes and wetlands. Hasenstab (1991) points out a problem in dealing with sites that are within wetlands (the same would apply to those within lakes or patches of organic soils). That is, if lakes and wetlands are deemed to have low probability of containing archaeological sites, a model will do a poor job of predicting sites that are even slightly within their margins. Thus distance to water/wetlands may not be as good a variable as distance to the edge of water/wetlands. This allows the center of a large lake to have a low probability for sites, while the edge has a high probability. Therefore, when the edge of a resource is the expected location, distance to edge measures were used.

When sites could be expected to be anywhere within or near a key resource (such as woodlands), a simple distance to the resource is measured. The same is true if sites are expected to be primarily within a resource (such as well-drained soils). In these cases, cells within the resource have values of zero and cells near the resource have very small values. This, in effect, is treating the boundary as being imprecise or variable. If a site is actually within a polygon of well-drained soil, but because of a mapping error falls just to the outside of that soil polygon, its low distance value will still contribute appropriately to the model.

Return to Top

6.2.2 Terrain

Digital elevation data consist of elevation values for points spaced at regular intervals in a grid or lattice. GIS software can derive all other characterizations of terrain from these data. It can also derive contour maps from the lattice of values.

The elevation data used for these models were at 30-meter resolution, i.e. 30 meters between data points. All elevation grids and layers derived in this project have vertical units in feet. Although contour maps can not be used for analysis in GIS, they are useful for display of data because they are familiar and interpretable by many people. Contour maps were generated for display purposes only. The following variables were used to characterize terrain:

6.2.2.1 Variables Derived from Elevation

Elevation: Elevation, measured in feet, was taken from one of two sources, depending on the best available data for any given area (Figure 4.2). Where available, 7.5 minute Digital Elevation Models (DEMs) (see Sections 4.5.1.3, B.6.3.1, and B.6.3.2 ) are the source of the elevation grids. For places where no 7.5 minute DEMs are available, elevation data from the statewide MGC100 database (derived from 1:250,000 DEMs, see Appendix B.3.2) were resampled to 30-meter resolution. Gaps between the merged quads were filled in by interpolation.

Height Above Surroundings: Archaeological sites tend to be somewhat higher than their surroundings. Higher land promotes drainage, provides a better view of the surroundings, and makes the position more defensible. This variable measures the height in feet above the lowest cell within 90 meters. Negative values indicate cells that are lower than their surroundings.

Prevailing Orientation: This variable is a function of aspect. Prevailing orientation measures the orientation towards south in degrees. The lower the value, the less the deviation from south. A south-facing slope would have a value of zero; a north-facing slope would have a value of 180.

Relative Elevation: The maximum vertical elevation change within 90 meters, measured as the absolute value in feet of the difference between the elevation of the cell and the elevation of the highest or lowest cell within 90 meters, whichever is largest.

Slope: This variable represents the maximum rate of elevation change over the 30 m distance between each cell and its neighbors, measured in degrees. Cells with very high values are on steeper slopes; cells with low values are on flatter land.

Solar Insolation: This variable was used only in Phase 2 to measure surface illumination, taking into account the direction to the sun (azimuth), height of the sun above the horizon (solar altitude), and the effects of both incident light and shading. This value may be derived for any day of the year, at any time of day. For this project, solar insolation is measured at noon on the shortest day of the year (December 21 or winter solstice). Places with higher values are receiving more solar energy and hence would be warmer than places with lower values.

Surface Roughness: This dimensionless roughness index is derived from elevation, slope, and relative elevation using weights and constants derived by Hammer (1993).

6.2.2.2 Variables Derived from Watersheds

These variables were introduced in Phase 3 to measure the size of major and minor watersheds in square meters, and distances to major and minor nearest ridges or divides in meters. The 81 major watersheds are consistent with the United States Geological Survey's 8-digit Hydrologic Cataloguing Unit (HUC).

The minor watersheds are subdivisions within the major watersheds.

Distance to ridges (water divides) provides a measure of territory accessibility, as paths often followed these features. Size of a watershed served as a measure of the degree of development a river system has evolved and, indirectly, of its relative age. Presumably, large watersheds would contain a wider variety and greater quantity of resources than small watersheds.

Distance to Nearest Major Ridge or Divide: This variable measures distances to major watershed (area of land enclosed by a continuous height-of-land drainage divide) boundaries in meters.

Distance to Nearest Minor Ridge or Divide: This variable measures distances to minor watershed (area of land enclosed by a continuous height-of-land drainage divide) boundaries in meters.

Size of Major Watershed: This variable measures areas of major watersheds, delineated by the Mn/DNR, in thousandths of square meters. Because these values were so large, they could not be stored in ARC/INFO grids as square meters.

Size of Minor Watershed: This variable measures areas of minor watersheds, delineated by the Mn/DNR, in square meters.

Because values for both watershed sizes and distances to ridges could be very large, both were transformed to square roots for Phase 3 modeling (see Chapter 7). Therefore, values reported in Chapter 8 are the square roots of the original distance or area measurements.

Return to Top

6.2.3 Surface Hydrology

Water and wetlands are important resources for drinking, cooking, and washing water, wildlife, fire protection, and transportation. Maps of modern water bodies may not reflect past conditions. Lake levels have fluctuated up and down over the past 10,000 years. Dams have been constructed, enlarging lakes or transforming streams into lakes. Consequently, some sites are in areas that are now covered with water. Some features that are now wetlands may once have been lakes. Wetlands and lakes that have been drained by natural or anthropogenic causes in the past 100 years may now be represented by remnant patches of organic or hydric soils. For practical reasons, the assumption was made that contemporary features could serve as surrogates for features from the past. This assumption is most likely to work for modeling relatively recent sites. Most recorded sites in Minnesota are thought to be relatively recent.

Several contemporary sources of information were available for surface waters. The National Wetlands Inventory (NWI) was available in digital format for the entire state from the beginning of this project. This database provided information on all water bodies that could be displayed as polygon features. It distinguished between lakes, rivers, and wetlands. NWI also distinguished between different kinds of wetland environments. NWI classifies lakes, rivers and wetlands by seasonality, making distinctions between permanent and seasonal water bodies. NWI was an excellent source of data and allowed for the analysis of various relationships between archaeological sites and water. However, it does rely on aerial photographs in its classification, without extensive field checking. Consequently, errors of misclassification can be expected. Moreover, this data source does not indicate the full extent of drained lakes, which are numerous in Minnesota.

The presence of artificial lakes and wetlands in the NWI coverages complicated data extraction and prevented a number of minor lakes and river reaches from being included in modeling. To partially remedy the situation, artificial lakes and wetlands, as coded in NWI, were removed from the data, and then selectively restored before modeling. In cases where major lakes had been created or enlarged by damming or mining, they were replaced with best approximations of the original lake or stream from historic sources, when such data were available. Most often this source was a Trygg map (Trygg 1964-1969). Trygg maps were used to delineate some lakes, e.g. Lake Winnibigoshish (between Itasca and Cass counties) and Mud Lake (in northwestern MN). In one case, a pre-dam shoreline was taken from an old USGS quadrangle. County boundaries could sometimes be used as surrogates for old stream courses. These reservoir lakes and streams were not replaced in the raw data layers (i.e., the Minnesota Wetlands Inventory), but only in the grids that were used for deriving variables in Phases 2 and 3.

The MnDOT base map, digitized from the USGS 7.5 minute quadrangles, was the source for single line perennial and intermittent streams. Lakes and double line rivers were not built as polygons in this database, and therefore could not be used. These were taken instead from the National Wetlands Inventory. For modeling, the NWI double line rivers and MnDOT streams were combined into a single grid.

6.2.3.1 Variables Derived from Surface Hydrology

Direction to Water or Wetlands: Direction to water was important for human settlements. Prevailing winds in Minnesota, particularly during the fire season, come from the southwest. Water bodies probably served as fire breaks, so locations to the east and north of water would be protected. Not only would these locations be safer for people, they may also provide food, fuel, and shelter in the form of trees that might not grow in the more fire-prone areas south and west of the water. Direction was measured as the azimuth. The azimuth is measured in degrees, where 0^o is the azimuth of the source cell, 90^o= east, 180^o = south, 270^o = west, and 360^o = north. For modeling, the sine of the azimuth was used. The sine of the azimuth will be 0 for east and west directions, 1 for north, and -1 for south. Consequently, a positive coefficient for direction means a higher probability that features are in a northerly direction (i.e. sites are south of lakes). A negative coefficient indicates a higher probability for features in a southerly direction (i.e. sites are north of lakes). Direct azimuth measurements and cosines of azimuths were also used in Phase 2 modeling, but appeared in many fewer models than the sine of the azimuth, so were discarded in Phase 3.

Distance to Water: Proximity to water is a vital consideration for settlement location. Distances to water were measured in meters. Sites, particularly villages and campsites, are expected to be close to shorelines. Lake edges were important for fishing and wild rice harvesting. Edges are easily detected by the GIS software. Distances from them can be measured away from the lake, to model sites at higher elevations than the modern lakeshore, or into the lake, to model sites that are now under water. However, edges were not relevant to the perennial and intermittent streams which are depicted as linear features, only one cell wide. A number of variables measuring distance to water were used in Phase 2 and 3 (see Table 6.1). Because their values could be very large, their square roots were used for modeling, and Chapter 8 reports the values of the square roots, rather than the original variables.

Distance to Wetlands: Wetlands are important for the rich sources of food they provide. Swamps also provide wood for fuel and may have been places to hide from enemies. Moreover, areas that today are wetlands may have been lakes under a wetter or cooler climatic episode. Distances to wetlands were measured in meters. When edges of wetlands were derived, the edges they shared with lakes and rivers were eliminated so that lake and river edges were not represented twice in the analysis. Thus, edges of lakes and rivers may be adjacent to wetlands or dry land, but edges of wetlands are defined as adjacent only to dry land. Several variables measuring distances to wetlands were used in Phases 2 and 3 (Table 6.1). Because their values could be very large, their square roots were used for modeling, and Chapter 8 reports the values of the square roots, rather than the original variables.

Composite Distance to Water: In addition to calculating distances to individual types of water (lakes, wetlands, streams) or surrogates for former water bodies (organic soils, Section 6.2.4.1) separately, composite variables, measured in meters, were derived from both surface hydrology and soils (Table 6.1). Because their values could be very large, their square roots were used for modeling, and Chapter 8 reports the values of the square roots, rather than the original variables.

Lake or Wetland Size and Seasonality: Large lakes and wetlands may have been more important resources than smaller features. Permanent water bodies may be more reliable resources than seasonal ones. Consequently, distance variables were defined for both the general category (distance to lakes) and for the special category (distance to large lakes, distance to permanent lakes). For the purposes of this study, large water bodies were arbitrarily defined as those larger than 120 acres (see Cowardin, 1977 and Section B.4.2.2). The sizes of the nearest water bodies of various types were measured in square meters and considered as separate variables in Phases 2 and 3. Both size and distance measures were transformed to square roots for modeling and are reported as such in Chapter 8.

Distance to Confluences: River and stream confluences were important nodes on transportation routes. However, the hydrologic data available do not indicate which streams are navigable or were in the past. Confluences were therefore identified only where streams of different classes met. A single aggregated variable, distance to nearest confluence between perennial or intermittent streams and large rivers, remained among the Phase 3 variables. Distances to confluences were measured in meters. Square roots of these distances are reported in Chapter 8.

Distance to Inlets/Outlets: Inlets and outlets are places where streams enter or leave lakes or wetlands. These may have been important transportation nodes. In some parts of the state, they may have been wild rice sites. Distances were measured in meters, then transformed to square roots for modeling.

6.2.3.2 Variables Derived from Elevation and Surface Hydrology

Vertical Distance to Water: A number of river valleys in Minnesota were carved by large rivers fed by glaciers. Much smaller rivers flowing at much lower elevations than their surroundings now occupy these valleys. From the perspective of the top of a bluff, the river may be near, but the vertical distance an obstacle. For this reason, vertical distance to water and vertical distance to permanent water, measured in feet, were included as variables. These were derived from a combination of hydrologic and elevation data. Water bodies included both lakes and streams.

Return to Top

6.2.4 Soils

Soils data were available in digital format from two sources. The MGC100 database contained a number of soil layers digitized from the state soil atlas (University of Minnesota). The source resolution of these layers was quite coarse (40 acres, with a source scale of 1:250,000), but they were available for the entire state. Because of the source scale, assumptions about precise locations of occurrences of soil types cannot be made. Perhaps the best interpretation of the MGC100 data is that it indicates areas where certain kinds of soil are prevalent. These data were used in the basic models because the database was available statewide.

The second source of soil data, county soil surveys, had been digitized for 48 counties (Figure 6.1) at the time this report was written. Most of these are in raster (EPPL7) format. Only 27 of these were available for incorporation into models during Phase 2 of the project (Table 6.2). These digital soil surveys are referred to as the "high resolution soils data" because they are digitized from source maps scaled at 1:24,000 or less. They vary widely in quality (Governor’s Council on Geographic Information, 1997). Some are considered to be out of date because they are based on a soil classification no longer acceptable by soil scientists. Some were mapped on rectified, but not orthogonalized, air photos that contain spatial errors due to terrain displacement. These may not register accurately to other GIS layers, particular in areas with significant vertical relief. Plans for improving digital soil data for Minnesota are discussed in Section 10.2.1.2.

6.2.4.1 Variables Derived from MGC100 (low resolution) Soils Data

Organic Soils: Organic soils were considered to be potential indicators of lakes or wetlands that had been drained or filled by sedimentation. Organic soils (those with 20% or more organic matter) tend to develop in wet (hydric) conditions. Thus the measurements of distances to edge of nearest area of organic soils and distances to edge of nearest large area of organic soils, in meters, are treated in the models the same as distances to edges of modern water bodies.

Drainage: Soil drainage may be important for site location. Dry, well-drained areas are more suitable for prolonged human habitation than are wet or seasonally wet areas. Well-drained soils were identified from the classification of soil types. The variable distances to well-drained soils was then calculated in meters and used as a model variable both in Phases 2 and 3 (expressed as square roots). The simple distribution of well-drained soils was used in Phase 2 to define the variable soil drainage by assigning a value of 1 to well-drained soils and 0 to all other soils. However, the scale of the source data would prevent the identification of inclusions of well-drained soil in otherwise wet areas or, conversely, of inclusions of poorly drained soils in otherwise well-drained locations.

6.2.4.2 Variables Derived from Surface Hydrology and Soils

Organic soils are included in composite variables with other water bodies, to indicate a more general "distance to edge of water". These variables measured distances to edges of nearest lakes, wetlands, areas of organic soils, or streams. All distances are measured in meters and transformed to square roots for Phase 3 modeling.

6.2.4.3 Variables Derived from High Resolution Soils Data (County Soil Surveys)

Because the digital county soil surveys were not available statewide, they could not be used to build statewide models. Variables derived from these data were considered in the Phase 2 enhanced models for portions of selected regions.

The counties for which high resolution soils data were acquired are listed in Table 6.2. However, Pennington and Wilkin Counties’ data could not be used for enhanced models because these counties contained too few known archaeological sites. The soil survey for Beltrami County covered only the southern half of the county and even then had large areas containing no data.

Table 6.2: Counties with High Resolution Digital Soil Surveys Available in Phase 2

Beltrami	Goodhue	Pennington
Big Stone	Houston	Pipestone
Blue Earth	Jackson	Redwood
Brown	Kandiyohi	Rock
Carver	Lyon	Stearns
Chippewa	Martin	Steele
Chisago	Murray	Swift
Dodge	Nicollet	Wilkin
Faribault	Olmsted	Winona

The high resolution digital soil maps for these counties had only rudimentary attributes, usually simply a soil map unit code. To obtain information about the soil characteristics, the soil value attribute table was linked to the Natural Resource Conservation Service’s (NRCS) statewide soils database. This database was provided as ASCII files. Data that could be related to the mapped soil polygons were extracted only after considerable manipulation of the database (see Section B.4.2.2). Problems included creating single records from multiple records for single map units, creating single items from multiple items, replacing missing values with reasonable estimates, adding variables and assigning their values, and extracting the records for a county and joining to the county soils attribute table. The variables derived are not listed in Table 6.1, since they could not be used for statewide modeling, but are discussed below and in Section B.5.3.

Soil Surface Characteristics: Analysis of soil characteristics focused on surface layers because it was assumed that pre-agricultural people interacted more with the surface of the soil than with the lower layers. Clearly some information is lost by using only surface data. Permeability of the subsurface may limit soil drainage to the extent that the surface layer is not well drained, though it might have a high permeability rating. Moreover, some surface characteristics vary between sequences within each surface map unit. Quantitative variables are expressed in the NRCS database as ranges of values for each soil map unit. These soil characteristics were generalized for analysis. Mean values were calculated for ranges and/or sequences for the same map unit. Soil surface characteristics that were used as variables in the enhanced models were mean depth to the lower boundary of the surface layer, mean clay content, mean available water holding capacity, mean organic matter content, mean soil reaction (pH), and mean permeability rate. Additionally, a subjective scale of suitability for archaeological sites was developed based on the soil textural classes. Low values on this scale were assigned to textural classes that were considered to have low potential for archaeological sites and high values to those considered to have high potential. These values were assigned intuitively by the project archaeologist.

Hydric Soils: Hydric soils are soils that are wet for prolonged periods, but do not necessarily have a high organic matter content. MGC100 provides a coarse resolution map of organic soils, but not all of these may still be considered hydric. County soil surveys allow the identification of hydric soils, not all of which are organic. Because of the source data, hydric soils are potentially at a finer spatial resolution, thereby present a more accurate representation of where previously existing lakes and wetlands were found than do the coarse resolution organic soils. Since variables derived from MGC100 organic soils were significant contributors to the Phase 2 basic models, the equivalent variables were derived using hydric soils as a source for the enhanced models. These variables were distance to edge of nearest hydric soils, distance to edge of nearest large area of hydric soils, distance to edge of nearest lake, wetland, or hydric soils, and distance to edge of lakes, wetlands, hydric soils, or streams.

Return to Top

6.2.5 Vegetation

Vegetation was an integral part of the hunger-gatherer’s landscape. Trees and shrubs provided food, fuel, building materials, and protection from the elements. Oaks were of particular importance because of the acorns they provided for food. Prairies were significant for hunting large game, while waterfowl concentrated in and around wetlands. Some vegetation resources were available only seasonally, others year-round. High local diversity of vegetation types would have made a wider range of resources easily accessible.

The distribution of vegetation types within Minnesota changed over the last 10,000 years as climate changed and can be reconstructed on the basis of palynological data and paleoenvironment models (see Section 6.3.3). Euroamerican settlement resulted in rapid, extensive forest clearance and the breaking of the prairie sod. Consequently, modern vegetation distributions do not reflect the resources available to the precontact population. Fortunately, the survey conducted to establish the Public Land Survey System (PLSS) recorded information about vegetation and species distributions in the early stages of Euroamerican settlement (1834 in southeastern Minnesota through the early 20^th century in the north). These surveys reflect the landscape after the contact period when some white settlers and economic activities, such as logging, were already present, but before large numbers of settlers arrived. Several sources containing components of the PLSS data were available for this project. These are the Trygg and Marschner maps and the DNR bearing tree coverages, all discussed below.

6.2.5.1 Marschner Vegetation Map

The best statewide source of digital vegetation data is the Marschner map (Marschner 1974). Francis Marschner, a U.S. Forest Service employee, compiled a map in the 1930s of the vegetation of Minnesota from PLSS records. Marschner’s map was drawn at a scale of 1:500,000, so it is a generalization of the surveyors’ records. It is not detailed enough to depict the patterning of vegetation types within the landscape, but is an indicator of areas where certain types were prevalent. Marschner’s mapping methods were not clear (Heinselman 1974). His polygons bear some resemblance to those on the surveyor’s plat maps, but he did not copy the plat maps faithfully. He apparently made reference to the bearing trees recorded in the surveyors’ notes, but it is unclear whether he also referred to species listed in line notes. He has classified forest types based on their species composition. However, his vegetation classifications are not always consistent with the surveyors’ descriptions of vegetation formations in the line notes.

The scale of the Marschner map, which results in a high degree of generalization, and the uncertainty associated with his methods for distinguishing between vegetation types make this source inappropriate for landscape scale analysis. Moreover, when this version was digitized, modern lake boundaries were added, reducing its reliability as an historic source even further. However, as it is the best available digital source of data, a number of variables (described below) were derived from it. These variables were used both for Phase 2 and 3 models unless exceptions are noted. There is some overlap between the vegetation variables used. For example, Big Woods, river bottom forest, and aspen-birch vegetation types are all considered separately as well as in the composite variable distance to hardwoods.

All of these vegetation variables are based on proximity measures. However, whereas sites are expected to be near, but not within, water bodies, sites could be expected either within or near vegetation communities. For that reason, distances to different vegetation types are not measured from the edges of these features. Sites anywhere within a vegetation type have a distance value of 0, while sites near a vegetation type have low values. All distances were measured in meters.

6.2.5.2 Variables Derived from the Marschner Map

Distance to Aspen-Birch: Aspen-birch includes both categories (hardwood and conifer) of aspen-birch mapped by Marschner. In both cases, the forests were dominated by quaking or bigtooth aspen and paper birch. These were assumed to be a successional stage leading to either Big Woods or conifer forest. Components of these forest types were present in a successional understory or were co-dominants in the canopy (Heinselman 1974). Aspen-birch may be an indication of a previous fire. Food products and fuel may have been found in this community, and paper birch was an important source of materials for making containers, canoes, and other necessities.

Distance to Big Woods: Big Woods is the term used in Minnesota for mesic deciduous forest. It was dominated by sugar maple, basswood and elm, with oaks, hickory, walnut, cherry, and potentially many other species (Heinselman 1974). One key characteristic of this vegetation type is that, to develop fully, it requires protection from fire. Thus, Big Woods vegetation would be expected to be in places protected from fires by high humidity or rainfall, moist soils, or natural firebreaks. Big Woods would have provided hunter-gatherers with food, fuel, and shelter.

Distance to Brushlands: Brushlands included brush prairie, aspen-oak land, and oak openings. These were all fire-maintained and usually occurred in the ecocline between Prairie and Big Woods (Heinselman 1974). They may have provided fuel and populations of browsing animals, such as deer, for hunting.

Distance to Conifers: This variable measures the distance to the nearest community containing coniferous trees. The communities included mixed hardwood and pine, white pine, white and Norway pine, Jack pine barrens and openings, pine flats, and aspen-birch (conifer phase). These are concentrated in the northeastern part of the state.

Distance to Hardwoods: Hardwood forest types included Big Woods, River Bottom Forest, and Aspen-Birch (hardwood phase). This variable, measured in meters, was included to determine if hardwoods in general were important, rather than specific hardwood communities. Oak openings were excluded because they typically were composed of groves or scattered trees rather than forest.

Distance to Mixed Hardwoods and Pines: This mixed forest occurs in the north-central and northeast parts of the state and is transitional between Big Woods and white pine forest (Heinselman, 1974). Presumably, resources from both hardwood and conifer communities would have been present.

Distance to Oak Woodland: Marschner called this community oak openings and barrens. This was a fire-maintained community that typically served as a buffer between prairie and Big Woods (Heinselman 1974). Rather than the continuous canopy of a forest, oaks were scattered individually or found in scattered groves. The understory may have been shrubby or grassy. Resources would have included wood and acorns, as well as habitat for grazing and browsing animals.

Distance to Pine Barrens or Flats: These included jack pine barrens and openings and pine flats. Jack pine barrens occurred on sandy glacial outwash or on thin rock outcrop soils (Heinselman 1974). It was a fire maintained mosaic of jack pine stands and nearly treeless heaths. Pine flats were mapped only near Duluth, and may have contained a significant amount of hemlock.

Distance to Pine Forests: Pine groves included both the white pine and the white and Norway pine forests on which the timber industry was built. The nearly pure white pine groves were old-growth white pine forests similar to those found in New England, Pennsylvania, Michigan, and Wisconsin (Heinselman 1974). These were much less extensive in Minnesota. Old growth white and Norway pine forests were more common. Immature pine stands generally were classified as jack pine barrens or aspen-birch (conifer). These were not included in the definition of this variable. This variable was used only for the Phase 2 models.

Distance to Prairie: Minnesota prairies were primarily of the tall grass prairie type (Heinselman 1974). They were not ideal human habitat. Prairies dominate the southwestern part of the state where higher temperatures and lower rainfall contribute to the fire hazard. The general absence of trees on these broad, flat uplands is attributable to frequent fire, rather than low rainfall per se. Large herds of grazing animals and some food products were found on the prairies. However, in the prairie regions, hunter-gatherers are believed to have congregated near water bodies where both water and wood were available.

Distance to River Bottom Forest: Marschner used this type throughout the state for forested floodplains and river valley bottoms (Heinselman 1974). Typically these are dominated by hardwoods that can tolerate saturated soils for some period of time. These include cottonwood, ash, elm, box elder, and hackberry. In the prairie regions, river bottom forests may be the only source of wood for fuel.

Distance to Woodlands: This measured the distance to the nearest woody vegetation, not including swamps. The vegetation communities included aspen-oak land, oak openings and barrens, Big Woods, river-bottom forest, aspen-birch (hardwoods and conifers), mixed hardwood and pine, white pine, white and Norway pine, jack pine barrens and openings, and pine flats. Presumably this variable would be significant in landscapes where more than one woody formation is present and there is no distinction between them in the resources they provided. This variable was not used for the Phase 3 models.

Vegetation Diversity within 0.5 km: This variable was defined as the "number of different vegetation types within 510 meters" of each cell in the grid. High diversity values indicate a greater variety of resources in the immediate vicinity. Ecological theory suggests that contact zones (ecotones) between different vegetation types should be rich in resources simply because of the larger variety of species present. For instance, people living near the boundary between woodlands and grasslands would have both forest and grassland game in hunting distance. Thus, diverse landscapes should provide a rich resource base. It is assumed that prehistoric humans, striving for efficiency, would have located within easy walking distance of their most important resources. This variable was dropped from the Phase 3 analysis because the variable vegetation diversity within 1 km performed better. This is probably attributable to the latter variable’s wider range of values.

Vegetation Diversity within 1 km: This variable is defined as the "number of different vegetation types within 990 meters" of each cell in the grid. High values indicate a variety of resources within a larger, but still easily accessible, area.

6.2.5.3 Trygg Maps

In the 1950s and 1960s, William Trygg transcribed the PLSS plat maps for the entire state to a scale of 1:250,000 (Trygg 1964-1969). These maps show natural features, such as lakes, streams, marshes, swamps, and prairies, and also cultural features, such as roads, trails, Indian villages, and settlers' cabins. They contain much of the spatial detail of the PLSS records that is omitted by Marschner. Because of the map scale, the locational accuracy and resolution are not great. However, they are more accurate and more suitable for landscape scale analysis than is the Marschner map. In Phase 2 of this project, vegetation and cultural features from these maps were digitized for 20 counties (Figure 6.2).

These data have serious limitations. Surveyors only mapped what they could see from their survey lines, which followed section lines in a one-mile square grid. Therefore, the maps show quite a bit of detail along the section lines, while the interiors of the sections are constructed mostly by conjecture. Many Minnesota surveys were conducted in the winter, so that straight lines could be surveyed across frozen lakes. This may have affected how well vegetation could be observed. Furthermore, some surveyors mapped quite a bit of detail, while others mapped only the minimum required. Finally, fraud and bias in the surveys have been documented (Bourdo 1956: 759), and surveyors may never have visited some areas at all.

Vegetation classes depicted on these plat maps are very broad. Prairies, marshes, and swamps are labeled and are easily interpreted. Bottoms are also labeled, but distinctions are seldom made between grassy bottoms and those that are wooded. Forest types are seldom distinguished. Experience with the original plat maps and surveyor’s notes indicate that all white, unlabeled areas on the Trygg maps were probably wooded. However, determining whether the woods are open savanna or closed forest, deciduous or coniferous, requires referring to the surveyor’s notes.

PLSS streams mapped by Trygg were not digitized for this project, as they were not accurately represented or complete (only streams that crossed section lines were represented). It was assumed that modern streams and topography were an acceptable indicator of historic stream locations. The exception was where modern reservoir lakes were encountered. These were removed from the NWI lake layer and sometimes replaced by streams digitized from Trygg maps. Other features that were only occasionally encountered were omitted if they did not have any apparent relevance to the model. The focus was on recording the cultural features, roads and trails, vegetation, and wild rice sites.

Because of time and budget constraints, the entire state could not be digitized. A sample of 20 counties was digitized to evaluate of the contribution of the data to the model (Figure 6.2). For these counties, several variables were derived for use in developing Phase 2 Trygg enhanced models. The variables pertaining to vegetation are in Section 6.2.5.4. The variables pertaining to cultural features are described in Section 6.2.6. None of these variables could be used for statewide models in Phases 2 or 3.

6.2.5.4 Vegetation Variables Derived from Trygg Maps

Distance to Grassland: Grasslands were an important resource, particularly for the large game they supported. However, they tended to be farther from water than wooded areas and to be fire prone. Visibility would have been higher in grasslands, so they were not as defensible as woodlands. Finally, in grasslands, people would have been more exposed to wind, precipitation, and fire.

Distance to Woodland: Wooded areas provided fuel, building materials, nuts and berries, game, and shelter from the weather and from enemies. Trees also tended to grow in places that were protected from fires and near water, another key resource.

Vegetation Diversity within 0.5 km: This variable was defined as the "number of different vegetation types within 510 meters" of each cell in the grid. High diversity values indicate a greater variety of resources in the immediate vicinity. The differences between this variable and the Marschner variable of the same name are a function of source scale and the number of vegetation classes identified by each source. The Trygg maps, at 1:250,000 scale, show some features (small patches of prairie or wetlands, for instance) that are not found on the Marschner map. These would tend to increase the diversity values derived from the Trygg maps. However, the Trygg maps do not distinguish different types of woodland communities. This has the opposite tendency, reducing the Trygg diversity values.

Vegetation Diversity within 1 km: This variable is defined as the "number of different vegetation types within 990 meters" of each cell in the grid. High values indicate a variety of resources within a larger, but still easily accessible, area. Differences between this variable and the Marschner variable of the same name would be the same as for the previous variable.

6.2.5.5 Individual Species

Several individual species were particularly important in Minnesota. Sugar maple was used as a source of sugar. Seasonal camps were established near groves of maples to take advantage of this resource. Paper birch was the source of bark, which was used for a variety of purposes, including canoes and containers. Utilization of birch bark also resulted in seasonal camps. The Kentucky Coffee Tree may have been a food source. Its distribution within Minnesota may be largely due to planting by indigenous people (Sand et al. 1995), although there are no ethnographic references to its use. Cranberries were also an important seasonal food source. Only the distribution of highbush cranberries, but not bog cranberries, was available in digital form for this study. The distribution of all these species is known to have varied considerably over the past 10,000 years and probably also over the last 500 -1000 years. All available data sources represent species distributions from within only the past 165 years.

Distributions of sugar maple, paper birch, Kentucky coffee tree, and highbush cranberries were mapped as points (in some cases) and polygons (in others) by the Minnesota Department of Natural Resources from very coarse resolution source material (probably more than 1:1,000,000). Because of the scale of the available data, these were interpreted very loosely as areas where the species were prevalent in the recent past. Distances to all of these species, in meters, were considered in developing Phase 2 models.

Bearing tree data for birch and maple from the PLSS (United States Public Lands Survey System) records became available in digital form and replaced Phase 2 coarse scale data in Phase 3 of the project. Bearing trees were recorded at ½ mile intervals along section lines during the surveys. This source records more observations for each species than do the coarse resolution tree species distribution maps, and it does not generalize the distribution of species into polygons, which may include areas where the species is not present. However, of the species of interest, only paper birch and sugar maple were common enough to be recorded in abundance. Variables distance to sugar maple and distance to paper birch were derived from PLSS data and used in Phase 3 models. The variables distance to highbush cranberry and distance to Kentucky coffee tree were removed from the Phase 3 modeling because of the coarse resolution of their only available data source.

6.2.5.6 Variables Derived from Tree Species Distributions

Distance to Cranberries: Highbush cranberries are actually a shrub, not a tree. Moreover, they represent only one source of the berries that would have been available to hunter-gatherers. Bog cranberries were also important, but no digital source was available to show their distribution. Cranberries could not be used as bearing trees in the PLSS survey. Consequently, this variable was not used in Phase 3.

Distance to Kentucky Coffee Trees: This hardwood species produces large beans that could have been a valuable food source where they were available. However, the tree was not abundant in Minnesota and was found primarily in the southern part of the state. Because of its rarity, it was not used as a bearing tree, so was excluded from the Phase 3 analysis.

Distance to Paper Birch: Paper birch was the source of bark, which was used for a variety of purposes, including canoes and containers. Utilization of birch bark also resulted in seasonal camps. Distances to this plant community were measured in meters. This variable was used both for the Phase 2 and 3 models, but was derived from different data sources (see Section 6.2.5.5).

Distance to Sugar Maple: Sugar maple was used as a source of sugar. Seasonal camps were established near groves of maples to take advantage of this resource. Distances to this plant community were measured in meters. This variable was used for the Phase 2 and 3 models, but was derived from different data sources (see Section 6.2.5.5).

Return to Top

6.2.6 Historic Cultural Features

The Trygg maps recorded a wide variety of both Native American and European-American cultural features mapped by the PLSS (United States Public Lands Survey System) surveyors. These included roads, trails, settlers’ cabins, logging camps, Native American villages, sugar camps, wild rice sites, and more. Some of these are indications of where Native Americans were living in the early historic period and may be correlated with traditional, prehistoric land use. However, only those features that were visible from section lines are likely to have been recorded. Consequently, these data should be considered a sample, not a census, of historic cultural features. Moreover, they reflect features in place after considerable contact had occurred. Contact undoubtedly altered the locations of Native American trails and settlements by the establishment of trading posts, Indian agencies, and other Euro-American settlements where Indians could trade for resources.

Because digital coverage of this source was not available statewide, the variables derived from these features’ locations were used only in the Phase 2 Trygg enhanced models. All distances were measured in meters.

6.2.6.1 Variables Derived from Mapped Cultural Features

Distance to Native American Cultural Features: This variable measured distances to Native American villages, sugar camps, and other cultural features as a group. Distances to individual types of features were not measured because of the sparseness of the sample. This variable was used in Phase 2 models only.

Distance to Roads and Trails: Transportation routes tend to follow least-cost paths. Routes, once established, tend to be perpetuated. Consequently, it is expected that many settlers’ roads followed trails first established by the indigenous population. Roads and trails usually lead to key resources and may also be associated with camp sites or other activities. This variable was used in Phase 2 models only.

Distance to Intersections of Roads with Water: Lakes, wetlands, and streams were all important resources and, presumably, important destinations. Three variables were derived to measure distances to places where roads intersected with water features of all types, where roads intersected with lakes or wetlands, and where roads intersected with streams. This variable was used in Phase 2 models only.

Return to Top

6.2.7 Disturbance

Archaeological sites can be modified by a large number of natural and human factors. In fact, site identification is facilitated by various processes that expose soil surfaces, such as wind erosion, water erosion, and plowing. Other transformations, such as pedoturbation or soil mixing, can also alter the context of archaeological materials (Wood and Johnson 1978). Without considering these post-depositional processes, correct archaeological interpretations can be difficult or impossible. Extreme cases occur when cultural materials are completely removed from their archaeological contexts or original locations, common in riverine environments and during construction or mining activities. Reference layers for several kinds of disturbance factors were considered in the Phase 2 models. These layers were derived from the MGC100 database, at a source scales of 1:125,000 and 1:250,000.

6.2.7.1 Variables Derived from Disturbance Factors

Mine Pits or Dumps: Mine pits and dumps are places where precontact period sites are expected to have been completely destroyed by human activity. They are recorded on the 1:125,000 scale MGC100 quaternary geology layer. Large areas on the Mesabi Iron Range in Northern Minnesota have been subjected to open pit mining. This variable was used in Phase 2, but not in Phase 3. Instead, mined areas were excluded from consideration (masked out of the grids) in the Phase 3 models because of their altered topography and hydrology.

Susceptibility to Erosion by Water: Water transports sediment by either displacing it down to lower positions through the force of gravity or to lower energy environments. Erosion and transport occur when rain strikes a bare ground surface, displacing soil particles. If the area is sloped, displacements tend to occur down slope. Erosion is highest where slopes are steep, vegetation cover is sparse, rainfall is intense or water is channeled (Strahler 1975:413-419). If the grade is steep enough, erosion channels and draws begin to form, directing water and displacing ever larger objects (including artifacts) as the total water discharge increases. Once runoff enters streams and rivers, the potential for massive movements of sediment downstream increases exponentially. For example, the average amount of sediment carried by the Red River of the North past Grand Forks, North Dakota in the summer exceeds 1,620 tons (Harrison and Bluemle 1980). Along rivers and streams, erosion generally occurs on outside bends or high energy areas and deposition on the inside bends. It is along the outside bends of rivers and streams that destruction of archaeological sites is the greatest (Picha and Gregg 1993). Water erosion is also concentrated along the shorelines of lakes. The MGC100 layer used for this variable was developed by application of the Universal Soil Loss Equation to other MGC100 layers, with the results reported in only four classes. This variable was used in Phase 2 only.

Susceptibility to Sedimentation: Deposition (sedimentation) occurs during flooding, when sediment is deposited outside the river channel where stream flow or energy decreases enough to drop out sediment. Natural levees form along the banks of some rivers due to this process. Deposition is also concentrated along the shorelines of lakes and within lake beds. The MGC100 source layer for this variable classifies only two levels of sedimentation risk. This variable was used only in Phase 2.

Susceptibility to Wind Erosion: Unlike water, wind is not usually capable of moving large particles or objects any appreciable distance (Strahler 1975: 567-577). Only under conditions of extremely high wind speeds will anything heavier than a sand grain become airborne. Consequently, by removing small soil particles from the surface (deflation), wind is more likely to expose an archaeological site than to remove it. Many sites in the high plains are discovered in this manner (Frison 1978). However, by removing the surrounding soil, much of the archaeological context of the site is destroyed. In some cases, cultural material from multicomponent sites may be mixed together due to deflation. Susceptibility to wind erosion is a function of wind speed, topographic exposure to wind, soil particle size, soil moisture, and vegetative cover. Deposition of blown material may protect sites by burying them more deeply. Deposition by wind may result in loess or dune deposits in Minnesota. No digital data regarding wind deposits were available. In Phase 2 large areas of data were found to be missing throughout the state. For this reason, the variable had to be discarded in these Phase 2 regions and the variable could not be used in Phase 3.

Return to Top

6.3 PALEOENVIRONMENTAL VARIABLES

As Minnesota’s environment has experienced considerable change since the first human occupation (Section 6.3.1). Reconstructions of paleoenvironments are important for modeling locations of very old and buried sites. These sites are poorly represented in the archaeological database and are not likely to be highly correlated with present environmental conditions.

Three primary sources of digital data are available for such reconstructions. These are a statewide paleoclimate model (Section 6.3.2) run for this project, palynological data (Section 6.3.3), and the geomorphologic data (Section 6.3.4). The interpretation of these results as GIS layers attempts to consider the global view on climate change as a series of synchronized events while acknowledging local variation in Minnesota. It is precisely these global correlations that Anfinson and Wright (1990:215-216, 226-228) call into question. In essence, they believe that the effects and timing of global post-glacial climatic changes differ regionally. In addition numerous archaeologists have assumed, but not always demonstrated, that climatic changes have impacts on precontact peoples.

Although the results and interpretations of the paleoenvironment analysis are reported in detail here, none of these layers was used for modeling. The reasons for this are discussed below.

6.3.1 Regular Periodic Character of Global Climate Change vs. Irregular Regional Variations

Considerable global climate changes that had important effects on the evolution of environment and mankind occurred in the Quaternary period (Wright 1970, Wright 1899). Climate history inferred from pollen sequences, supported by fauna and geologic data, is clear in outline, though minor fluctuations and detailed regional variations exist (Watson 1991). Currently, a considerable degree of controversy remains regarding the periodical nature of past climate change.

One hypothesis is that a regular recurrent pattern of global climate change, with major cycles of aridity and humidity, happened concurrently throughout the Northern Hemisphere (Rychagov, 1984). Evidence indicates that marked global similarities in climatic regimes existed during the late Pleistocene and early to middle Holocene in the Old World and in North America (Watson 1991, Bowen at al. 1986). Numerous studies have been implemented to correlate glacial and post-glacial events in the Northern Hemisphere, beginning with establishing broad scale correlations (Wright 1899) and continuing with recent studies on absolute chronology. One of the serious and credible attempts to discover correlations between Quaternary glaciations in the Northern hemisphere is based on lithostratigraphical, thermoluminescence, potassium-argon, and fission-track dating obtained in Europe, Asia and North America. This study and a correlation chart published revealed a large degree of hemispheric correlation of glaciation events. However, the authors themselves point out the provisional character of the correlations and the general relationships that can be derived from them (Sibrava et al. 1986).

The pattern of postglacial climate change, reconstructed on the basis of data from Europe and Western Asia, is presented in studies as a regular periodic oscillation (Figure 6.3). According to paleoclimate and paleoenvironmental data for Eurasia and the Mediterranean Region, 1,850-1,000 year humidity-aridity cycles are typical. Each cycle starts with a cool and moist phase, increased mountain glaciation, increased runoff, and rising lake levels. The second phase has a warm, dry climate, decreased mountain glaciation, and lower lake levels. These cycles cause shifts up to 2-3 degrees latitude of climate and vegetation zones on the Plains (Rychagov 1984). Cold, dry steppe conditions prevailed in Europe and Western Asia until 11,000 B.P. After 11,000 B.P. warmer and moister conditions developed, allowing Mediterranean oak-savanna vegetation to expand from sheltered areas until approximately 5000-4000 yr. B.P. (Watson, 1991).

Other authors dispute this position, arguing that major global shifts in climate were less significant than regional variations. One component of climate change may cause shifts in other components. These secondary changes may not be of the same intensity everywhere, or even of the same direction. Some took place in cycles lasting only a few hundred years, and changes in adjacent areas were not necessarily either synchronous or of the same magnitude and direction (Wakefield 1970).

Serious consideration of the regional implications of climate change is very important for Minnesota. The state occupies a geographically intermediate position between the Great Plains Grassland, the Eastern Deciduous Woodlands, and the Northern Boreal Forest. Temperature, which determines major changes in vegetation zones, has very steep gradients over the state. Thus, minor inaccuracies in model data may result in significant shifts in assumed positioning of vegetation zones in the past.

Return to Top

6.3.2 Paleoclimate Model for Minnesota

As part of this project, the Bryson Paleoclimate Model was run for Minnesota for the last 12,000 years (refer to Appendix F for a discussion of the model). The model divides the post-glacial period into a number of climatic episodes and depicts rapid global changes caused by shifts in prevailing air mass patterns. Model inputs consisted of 30 years of temperature records from 123 Minnesota weather stations and precipitation records from 163 Minnesota weather stations. Output consisted of annual precipitation, mean annual temperature, summer precipitation, mean summer temperature, winter precipitation, and mean winter temperature at 200-year intervals back to 12,000 B.P. These output were provided in text files that also contained meteostation geographic coordinates.

For efficiency in converting the paleoclimate model into GIS layers, five dates in the last 12,000 years were selected. These are considered to be critical dates in the Late Glacial/Holocene. The following climatic/vegetation sequence developed through varve studies at Elk Lake in Itasca State Park (Bradbury and Dean 1993) provides a context for the selected dates:

Late Holocene	3500 B.P. to present
Middle Holocene - Late Holocene transition	4500 - 3500 B.P.
Middle Holocene	7800 - 4500 B.P.
Early Holocene - Middle Holocene transition	8500 - 7800 B.P.
Early Holocene	10,000 - 8,500 B.P.
Late Glacial - Early Holocene transition	11,000 - 10,000 B.P.
Late Glacial	11,600 - 11,000 B.P.

In this part of the state, 10,000 - 8000 B.P. is the postglacial period (coniferous forest); 8000 - 4000 B.P. the prairie period (oak savanna); and, 4000 B.P. to Present the mesic-forest (modern) period. The five dates selected, which have both archaeological and climatic/vegetation significance are:

1. 7000 B.C. (9000 B.P.): This date is solidly in the coniferous period (late Early Holocene) and is a good middle date for Late Paleoindian archaeological remains. It is a critical period for Late Paleoindian occupations in Minnesota. Late Paleoindian remains are much more abundant in the state than are Early Paleoindian remains, and many supposedly Early Paleoindian remains could date to the Late Paleoindian period.

2. 6000 B.C. (8000 B.P.): A date in the Early Holocene - Middle Holocene transition during which significant sedimentation and alluvial fan formation took place. Most of the best known early buried sites in the region (Rustad Quarry, Granite Falls, Cherokee Sewer, Itasca) date to this transitional period, when prairie was replacing earlier coniferous forests and runoff was high.

3. 4000 B.C. (6000 B.P.): A date in the middle of the prairie period near the peak of the most eastward expansion of prairie. This is during the Middle Archaic, an archaeological period about which very little is known, presumably because sites are buried, are at the bottom of later lakes, were fewer in number, or were smaller (e.g. formed by widely dispersed families of bison hunters). Compared to the active 6000 B.C. period, this is a "black hole" in Minnesota archaeology, but many sites should date to this period.

4. A.D. 1000 (1000 B.P.): A date in the "medieval warm period" when significant cultural changes were occurring, presumably because of population growth and political pressure from the south rather than adjustments to new environments. Big village sites begin to appear at this time. It also represents the ca. 1000 B.C. - A.D. 1300 time period to which most sites in the archaeological database belong.

5. A.D. 1400 (600 B.P.): A critical date in the Pacific climatic episode when many prairie lakes (such as Devil’s Lake in North Dakota) apparently dried up. Significant culture change in all parts of the state occurred after ca. A.D. 1250 - 1300. This included Oneota expansion to the south and settlement agglomeration and Sandy Lake pottery to the north. This is a period about which little is known compared to the 1000 B.C. - A.D. 1300 period.

7000 B.C., 4000 B.C., A.D. 1000, and A.D. 1400 represent the four main critical dates. These are periods that represent major stable climatic episodes (Early Holocene, Middle Holocene, Late Holocene, the Pacific climatic oscillation within the Late Holocene) and presumably stable patterns of settlement. Most sites in the SHPO archaeological database were probably adaptations to a Minnesota represented by the A.D. 1000 date. Most of the sites probably date between ca. 1000 B.C. and A.D. 1300. This was also the period of the "Little Ice Age", which is atypical of the Late Holocene. These more recent time slices (A.D. 1000, 1400) should be helpful in understanding the locations of the sites.

Other dates to consider are 2000 B.C. and A.D. 1700. 2000 B.C. is during the Middle Holocene - Late Holocene transition and the period when Archaic sites begin to become more visible in the archaeological record (e.g., Old Copper in the north and Oxbow points in the west and north). It represents the latter part of the Middle Archaic. If the emphasis is on finding earlier sites poorly represented in the archaeological database, then dates like 7000 B.C., 6000 B.C., and 2000 B.C. would be more critical than more recent dates (A.D. 1000 and A.D. 1400). More recent sites are more visible archaeologically and fall within a more or less "modern" environmental context (and thus require less paleoenvironmental modeling).

The Late Holocene climate also has been variable. A.D. 700 - 1200 was the unusually warm and moist "Medieval warm period". A.D. 1700 is representative of the "Little Ice Age," when it was colder and snowier in the state. It should be noted that the A.D. 1550-1821 Native American archaeological record is very poorly known in Minnesota. However, models based on archival records are probably a better source of site location for sites late in this period than is paleoenvironment modeling.

6.3.2.1 Spatial Interpolation of the Paleoclimate Model

ARC/INFO, version 7.1.2 software, developed by Environmental Systems Research Institute, Inc, was used to spatially interpolate paleoclimate data. Climate variables for the contemporary period and time slices 600, 1000, 6000, 8000 and 9000 years before present (B.P.) were mapped first as points. From the raw model output, averages from two 200-year intervals were calculated to represent each of the critical dates. For example, values for 8900 B.P and 9100 B.P were averaged to derive values assigned to 9000 B.P. The resulting text files were then edited to convert them into the necessary format and further processed in ArcView to create shape files, using the meteostation coordinates to generate points. These shape files were later transformed into point coverages in ARC/INFO. Surfaces were interpolated from these points using ARC/INFO GRID. The interpolation functions SPLINE, IDW and TREND were each tested. The preferred method for surface interpolation was selected by comparing the results of several surface interpolations of modern data with published climate maps for Minnesota (Borchert and Gustafson 1980).

TREND performs a trend interpolation on a point data set and fits one polynomial equation to the entire surface, detecting trends in the sample data. It is a fast way to show general trends in a surface (i.e. rather than modeling the surface precisely). By using a polynomial regression to fit a least-squares surface to the input points, it finds the best-fit equation to generate the entire surface. The surface it generates will rarely pass through the original data points. The order of the polynomial must be specified. The higher the order, the more complex the surface. A first-order polynomial creates a tilted, rather than curved, surface. A second order polynomial creates one curve (i.e. a ridge). A third order polynomial creates two curves, and so on. Ultimately TREND was rejected because, with the sparse spatial distribution of available point data, the generalization of interpolated surfaces was too coarse (Figure 6.4).

SPLINE uses a two-dimensional minimum curvature interpolator to create a surface that passes exactly through the sample points and minimizes certain aspects of surface curvature. SPLINE gave a reasonable approximation of surface contours as compared to the existing maps. However, in both TREND and SPLINE, a noticeable distortion of interpolated surfaces was observed at the state borders (Figure 6.4). Both techniques apparently assume the values beyond the borders to be zero because there are no other values represented there.

The Inverse Distance Weighted (IDW) method determines cell values using a linearly weighted combination of a set of sample points. The weight is a function of inverse distance. This method is most appropriate when sampling is sufficiently dense to represent the local variation to be simulated. IDW did not lead to peripheral distortions and gave a good extrapolation of contours beyond the state border. However, it produced many small circular shaped "eyes" around sparsely spaced stations within the state (Figure 6.5), producing a patchy pattern.

Ultimately, the SPLINE (two-dimensional minimum curvature interpolation) method was chosen. Distortions of surfaces near the border (SPLINE) were preferred over those in the middle of the state (IDW). To mitigate the problem of peripheral distortions, additional meteostations were added to the input climate coverage. Supplementary stations were positioned within 50 miles of the Minnesota border as "mirror" stations to those located within the state at the same distances. These stations prevented the most extreme distortions along the border (Figure 6.6). This method of interpolation (SPLINE + "mirror" stations) is presented here as a temporary solution of handling the available data. For future work with this model, it would be desirable to obtain modeled paleoclimate data for the neighboring states (Wisconsin, Iowa, North and South Dakotas) and Canada (Ontario and Manitoba provinces).

6.3.2.2 GIS Analysis of the Minnesota Paleoclimate Model

Temperature and precipitation surfaces for annual, summer and winter periods for all time slices under consideration were analyzed. To obtain relative values of past climate characteristics, with respect to present conditions, modern values were subtracted from values for each of the past time slices. The resulting average, maximum and minimum difference values were used to relate the GIS climate model results to known climate changes reconstructed by other studies. Charts were prepared to depict temporal changes in Holocene annual and seasonal precipitation and temperature (Figure 6.7a, 6.7b). The surfaces illustrate spatial changes of the modeled climate parameters for each time period under consideration (Figures 6.8, 6.9, 6.10, 6.11, 6.12, and 6.13).

Precipitation

For earlier time periods, annual precipitation values depicted by the modeled surfaces show an almost linear increase through time. Average annual precipitation increased from 9000 B.P. to 600 B.P. by 53 to 250 mm (Figure 6.8). The driest period, according to model results, was 9000 years ago. Slightly more moist conditions existed 1000 years ago as compared to 600 years ago. However, local variations were more significant, as differences in spatial patterns increased dramatically over time. Modeled mean annual precipitation 8000-9000 years ago was 500-600 mm less than present in some parts of Minnesota. However, along the eastern border of the state, modeled rainfall in the same period was only 60 mm less than at present (Figure 6.8).

Holocene precipitation changes, as depicted by this climate model, are not completely consistent with known paleoenvironmental data from paleontology and stratigraphy. The uncertainty of the results increases for more remote time slices. There are also contradictions between model results and current concepts of regional paleoclimate. The current assumption is that, during the last 10,000 years, average temperature and moisture regimes have changed from cool/moist (10,000-7500 B.P.), to warm/dry (7500-6000 B.P.), and back to cool/moist (6000-0 B.P.). Other estimates, based on fluvial stratigraphy from the Driftless Area of Wisconsin, indicate that, in the period of maximum warmth and dryness (7500-6000 B.P.), average annual stream runoff probably was 40-60 % less than at present (Knox 1981). This would suggest a similar pattern in southeastern Minnesota. In contrast, model data show the highest average annual precipitation in 6000 B.P., as compared to all other periods (Figure 6.8). The model also indicates higher summer precipitation in southeastern Minnesota in 6000 B.P. than in other periods (Figure 6.9). Furthermore, considerably lower annual average precipitation values are modeled in the periods earlier than 6000 B.P. This is in conflict with the concept of a cool and moist climate 10,000-7500 B.P.

The most consistent results from the climate model, in comparison to paleoenvironmental data, were obtained for the Pacific climatic episode (600 B.P.). This period is depicted as having lower temperatures than present, a greater extent of low precipitation values in the north-west corner of the state, and a smaller extent of high precipitation values in the east, as compared to present.

These results indicate that the climate model may be inaccurate for earlier periods. By examining the seasonal precipitation surfaces, it becomes apparent that drops in modeled annual precipitation in 6000, 8000 and 9000 B.P. are entirely a function of drops in the modeled winter precipitation (Figure 6.10). While summer rainfall remains, with minor fluctuations, at approximately the same level over all modeled time periods, winter precipitation is 5-6 times lower, in the earlier stages of the modeled time span (8000-9000 B.P.), as compared to the later stages. No published research results indicate any climate change of this magnitude in Minnesota. Most likely, the method of modeling winter precipitation is erroneous, producing too large a reduction of values as it continues back through time. Some meteostations have negative winter precipitation values for the periods 8000-9000 B.P., which is certainly incorrect.

Geological evidence of higher than usual sedimentation rates in Minnesota 8000 years ago suggest higher rainfall and runoff amounts at this time (Knox et al. 1981). However, the climate model does not reflect this. There are two possible explanations for the discrepancy. First, the low model values for winter precipitation, described above, may have reduced average annual precipitation values. Second, sporadic runoff, related to glacial lake surges and other fluvioglacial processes, may have contributed to sedimentation, yet not be reflected in the precipitation values for this period.

Two observations raise concerns about the accuracy of this model. First, precipitation data from the Bryson Paleoclimate Model for the contemporary period appear to have 20-30% higher annual precipitation values than do published U.S. Weather Bureau average data for the 1931-1960 period of record (Borchert, Gustafson, 1980). Modeled winter precipitation data, for the time slices 6000, 8000, and 9000 B.P., seem to be erroneous, reaching very low, sometimes negative, values.

Temperature

Model results indicate that annual temperature fluctuated over the period under consideration. Temperature increased rapidly between 9000 and 6000 B.P. The earliest periods modeled were marked by annual temperature 2-4 degrees C lower than at present (Figure 6.11). Most of the difference is attributable to winter temperatures (Figure 6.13). Summer temperatures were not significantly different from present (Figure 6.12). Small differences in temperatures were modeled in the northern part of the state. Southern and east-central areas, however, were considerably cooler 8000-9000 B.P. as compared to present conditions. The temperature surfaces show no considerable warming between 9000 and 8000 B.P. The model actually indicates slightly lower summer temperatures in central Minnesota in 8000 B.P., the opposite of the expected trend. Many other sources indicate abrupt environmental changes at this time due to glacial retreat. At the same time, prairies succeeded boreal forest, represented by spruce. These inconsistent results are further evidence of possible problems with the model.

The period 6000 B.P. was characterized by a more continental seasonal temperature pattern than at present. Summers were generally warmer and winters colder and drier. The summer temperature surface (Figure 6.12) shows a zone, with temperatures above 20 degrees C, spreading far beyond the southwestern half of the state. In all other periods this zone occupies about only one third of the state.

Surfaces for 6000 B.P., with their continental seasonal temperature pattern and lower annual rainfall than at present, are consistent with the peak of the most eastward expansion of prairies in Minnesota. Summer precipitation dynamics, however, do not clearly show the magnitude of aridity in 6000 B.P. On the contrary, the maximum summer precipitation and the westernmost advance of isohyets 450 mm and higher occurred, according to the model, at 6000 B.P. This is in a total disagreement with the existing evidence of the most eastward advance of prairies at this time.

The "medieval warm period," approximately 1000 B.P., is indicated on the surfaces by slightly higher winter temperatures in the southern and central parts of the state, compared to the preceding and subsequent periods. Summer temperatures at this time are higher than at 600 B.P. as well. Average annual temperatures were warmer by an average of 1 degree C between 1000 – 600 B.P. compared to the present (Figure 6.11), while regional differences for the same periods ranged from -3 to +2 degrees. However, the model indicates the warmest summers for 6000 B.P., not for this period, which is consistent with the prairie expansion at that time. Summer precipitation dynamics do not clearly show the increase in humidity in 1000 B.P. that is evidenced elsewhere.

As this analysis reveals, modeled temperature trends, except for the earliest periods of 8000-9000 B.P., are generally in agreement with other existing evidence of past climate change. However, seasonal precipitation values derived from the Bryson climate model do not correspond to known trends of paleoclimate change in Minnesota for most of the periods considered. This raises questions about the quality of the climate model results and their usefulness for archaeological predictive modeling.

6.3.2.3 Quality of Climate Model Results with Respect to Quaternary Geomorphic Events

Implementation of this paleoclimate model should be undertaken with caution. As discussed above, the reliability of model estimates and the surfaces generated from them decrease significantly for earlier time periods. Cataclysms associated with Lake Agassiz had important regional effects, modifying the impacts of global climate changes in ways the model cannot predict. The impact of Lake Agassiz and its overflow influenced environmental conditions in Minnesota from 10,900 to 8500 B.P. Shortly after 11,000 B.P., lower elevation outlets into the Superior basin were uncovered by retreating ice, and the level of the Lake Agassiz fell in a series of steps, as water poured eastward into Great Lakes. The resulting low water Moorhead Phase ended about 9900 B.P. as the Marquette glacial advance crossed the Superior basin at the Upper Peninsula of Michigan and again blocked Lake Agassiz’s eastern outlets. Overflow returned to the Mississippi River watershed briefly, but by about 9500 B.P. the eastern outlets into Superior basin had again become ice free, initiating the Nipigon phase. Water levels fell abruptly at this time and a series of catastrophic bursts, each of which released up to 4000 cubic km of water in less than a year or two, flooded through the Nipigon basin to Lake Superior. By 8500 B.P. ice had retreated far enough to allow Lake Agassiz to overflow across northern Ontario into Lake Ojibway, by-passing the Great Lakes (Teller 1985).

Dramatic regional climate and environmental changes continued. Around 5000 B.P. the Nipissing Great Lakes formed impressive shoreline features all around the basin. Some of these are now uplifted at least 30 meters in the northeast part of the state. Lake Superior was finally separated from Lake Huron - Michigan about 2000 years ago by continuing isostatic uplift of the Sault Ste. Marie threshold. Isobases of glaciostatic rebound in the Superior basin indicate 900-1000 feet of uplift after ice retreat. Finally, isostatic rebound appears to be continuing at the present time. Apparent vertical movements indicate the northeastern corner of the Superior basin rising 27 cm per century and the Duluth area subsiding 21 cm per century relative to Point Iroquois, which is the head of the St. Mary’s River (Farrand and Drexler 1985).

The conclusion from this review of geological events is directly related to the quality of modeled temperature and precipitation values used to generate the paleoclimate maps. For the earlier modeled periods (8000-9000 B.P.), the method of predicting the past from existing meteostations must be inaccurate. This method does not produce adequate results for many areas of Minnesota, where the configuration of large lakes changed through Holocene. The current position of stations along the north shore of Lake Superior imprints humidifying and moderating effects of this large water body on adjacent areas through all modeled periods. However, Lake Superior did not exist at its current shoreline for the considerable portion of the earliest period modeled. The lake experienced both transgressions and regressions due to retreats or advances of ice lobes, isostatic land surface movements, outlet changes, and Lake Agassiz surges. With these coastline changes, the territories experiencing moderating lake effects changed. On the other side of the state, inland stations at the northwestern corner of Minnesota impose contemporary continental features of dry and warm summers and cold winters through all periods modeled. This is inappropriate for the periods when Lake Agassiz was extant in the northwestern part of the state, from 10,900 to 8500 B.P. During this time the cooling and moderating effect of this large water body should extend eastward with the prevailing atmospheric circulation over all of north central Minnesota.

Improving the model results would require updating data for inland and coastline stations for the earlier time periods, consistent with the moderating and continental effects appropriate to known stages of glacial lakes Agassiz and Superior. Minor adjustments may be required for the earlier time periods in relation to the absolute elevation changes in northeastern Minnesota. An uplift of the magnitude of 800-900 ft (as reported by Farrand and Drexler 1985) occurred due to rebound of the land surface after the ice retreat.

6.3.2.4 Paleoclimatological Variables

There were no variables derived from these data for the Phase 3 models, as data quality did not meet all expectations. GIS derived temperature surfaces, especially those for the later periods (6000-600 B.P.), provided a reasonable illustration of known trends of past climate change. However, errors and inaccuracies were apparent in predicted values for the earlier periods (8000-9000 B.P.). The least reliable results were obtained from precipitation values. Representation of seasonal precipitation was poor with respect to known evidence of paleoclimate trends. Additional input data and model adjustment and refinement may improve the results and make these data suitable for contributing to future modeling.

Return to Top

6.3.3 Palynological Data for Minnesota

As complex computer models are being used to predict climatic changes and their effects, it is important that these models be accurate. Since such models are usually developed using knowledge of the current climate, one of the best ways of testing their output is to check how well they can replicate past climatic conditions. The only way to actually determine these conditions is by analyzing the records found in ice and sediment deposited in the past.

Pollen, primarily from sediment cores of lakes and bogs, has been used extensively to interpret the past distribution of vegetation. In Europe and North America, the density of the data is relatively high. These data are most reliable for the more recent past. The Holocene, the geologic epoch that encompasses the last 10,000 years, is well represented. Beyond 12,000 B.P. the reliability of radiocarbon dating to determine pollen age deteriorates, and beyond 18,000 B.P., when the extent of ice sheets during the last glaciation was at its maximum, there are very few data at all.

Many cores have been collected over the years in Minnesota, with some parts of northern Minnesota containing more than one core per county. This is beneficial for examining the east-west migration of the coniferous forests in the northeast quadrant of the state. The prevailing wind is westerly, transporting the pollen from the forest back into itself. Thus, pollen assemblages from this region should reflect local vegetation.

6.3.3.1 Methods

Pollen data for the upper Midwest were obtained from the National Geophysical Data Center (http://www.ngdc.noaa.gov/) in Boulder, Colorado. These data consisted of raw pollen counts (the number of pollen grains found at each depth), radiocarbon dated depths, and geographic coordinates. The core locations are lakes or bogs that do not correspond to the weather stations for which the climate data were modeled. The density of pollen data samples (core locations) was lower than for climate data (meteostations). In addition, cores were located in clusters, resulting in an uneven coverage of the state. East-central and northwest-central areas of the state are relatively well supplied with pollen data. In contrast, pollen data are almost totally absent from the northwest, southeast and southwest corners of the state. All species considered in the analysis have similar spatial and temporal distribution of pollen cores.

Birch pollen cores’ geographic distribution, (Figure 6.14) illustrates a non-uniform pattern of data typical for all species. Pollen samples were less abundant for earlier periods: Sixty-eight birch samples were available in Minnesota for the present time, but only 26 for 9000 B.P. Core locations for all time periods were generally the same, with minor variations. The number of pollen samples within Minnesota borders was limited, so additional samples from outside the state border were used for the interpolation. In most cases this allowed derivation of relatively smooth pollen surfaces. In some instances, however, it resulted in artifacts near the state border, where in-state data were absent and out-of-state single samples affected interpolation results.

These raw pollen counts and radiocarbon dated depths were processed into pollen percentages to facilitate comparison and analysis. However, this represents merely a compilation of the percentage of pollen grains of a genus in a specific core taken from a specific lake. It does not reflect the proportion of that genus in the local vegetation at that place and time. The pollen percentages and radiocarbon data were processed, using linear interpolation, into 100-year time slices for the last 18,000 years.

The results were provided in text files, one text file per species per time slice, with varying numbers of core location records represented in each file. ArcView was used to convert the text files into shape files. ARC/INFO was used to convert the shape files into point coverages. Because core sample locations are sparsely distributed, data for five 100 year periods each were combined to represent 9000 B.P., 8000 B.P. and 6000 B.P. Samples for three 100 year time periods were combined to represent 1000 B.P. and 600 B.P. Surface interpolation from point coverages representing each critical date was implemented in GRID (refer to the discussion of surface interpolation methods in Section 6.3.3). Because of the very low density and clustered distribution of the pollen cores, only a very coarse approximation of a pollen surface may be expected from these data. SPLINE and IDW methods of interpolation can produce inadvertent complexity of grid surfaces, which might be misleading. Consequently, the TREND method of surface interpolation was chosen as most appropriate for the pollen data (Figures 6.15, 6.16, 6.17, 6.18, 6.19, 6.20, 6.21). Charts depict the temporal dynamics of average, maximum and minimum values of pollen count for each species analyzed (Figures 6.22a, 6.22b, 6.22c).

6.3.3.2 Temporal Spatial Change of Vegetation in Minnesota in the Past 9000 Years

The known history of Minnesota flora includes several shifts in coniferous, deciduous, and prairie vegetation zones as a result of climate change (see Chapter 3 for a more extensive discussion). These transformations were abrupt and started about 9500 years ago. In the late Wisconsin period the climate change that accelerated the ice retreat also caused the spruce forest to disappear. In the western part of the state the spruce forest was replaced by prairie. In the eastern part of the state spruce forest yielded to birch in the south and to pine further north. Temperate deciduous trees then succeeded these forest communities. The temperate forest was soon interspersed with grassy openings, and by 8000 B.P. a fully developed prairie covered the region, extending to the north-east well beyond central Minnesota. Reversal of the climate trend about 7000-6000 B.P. caused the prairie to recede to the west, and groves of deciduous hardwoods reoccupied the slopes and depressions far out into the modern prairies (Wright 1970).

The pollen surfaces interpolated from the pollen data largely agree with the established history of Minnesota paleoenvironments, depicting known stages of climate and vegetation changes. Spruce is indicative of cold and moist climate conditions. Spruce pollen surfaces (Figure 6.15) show a maximum southern extent of the species in 9000 B.P., with an abrupt retreat to the northeast between 8000 and 6000 B.P. The spruce range increases again 600 B.P., concurrent with the "Little Ice Age" of the medieval period. Artifacts of non-uniform samples occur along the western border and in the southwestern corner of the pollen surfaces for periods between 8000 and 600 B.P. Pollen cores do not present within northwest and southwest areas of the state, but single cores with high pollen counts are found outside the state in proximity to its western borders. In this instance the interpolation of point data produced artificial zones of higher pollen percentage, filling empty areas along Minnesota's west and southwest borders.

Prairie species (i.e. sage), which are indicative of warm, dry conditions, show the opposite trend. Sage pollen surfaces (Figure 6.16) reveal the maximum east-central extent of prairies within the period from 8000 to 6000 B.P. and a retreat to the west in subsequent periods. Ragweed surfaces (Figure 6.17) show the same trend, though not so explicitly. However, though ragweed is inferior to sage as an indicator of prairie conditions, it is an excellent marker for European settlement. Ragweed pollen reaches its highest level at the present.

Oak pollen surfaces (Figure 6.18) illustrate another important climate trend, providing evidence of warmer and more humid conditions in Minnesota. Refuge areas (i.e., the driftless area in southwestern Minnesota) that sheltered temperate hardwood vegetation during cooler periods exhibit higher values of oak pollen for the earliest periods (8000-9000 B.P.). Oak savanna then expanded far to the north at the end of the prairie period (6000 B.P.) Temperate deciduous hardwoods, including oak, reached their greatest extent in Minnesota between 6000-1000 B.P., marking the gradual reforestation of prairies after 6000 B.P. and the "medieval warm period" of 1000 B.P.

Pine pollen surfaces (Figure 6.19) generally confirm the trends outlined above. Pine expands its range in colder periods. The maximum spread of pine in Minnesota was observed at 9000 B.P., with a second peak of expansion in 6000 B.P. During the last thousand years, pine distribution in Minnesota has been relatively stable and concentrated in the northeastern part of the state.

Birch pollen surfaces (Figure 6.20) show concentrations in the northeastern part of the state in all periods. Advances of birch to the south and west are apparent in cooler and moister periods (9000-8000 and 1000-600 B.P.). At present more birch is in the northwest than in previous periods. As birch is a component of secondary forest successions, the trend is likely evidence of logging and not of climate change.

Sedge dynamics in Minnesota over the last 6000 years show only moderate correspondence to the temporal dynamics of humid conditions. Low pollen percentages were observed 6000 B.P., which agrees with drier conditions at this time. Higher percentages of sedge pollen are observed in the southern part of the state between 1000 and 600 B.P. concurrent with the "Little Ice Age" of the medieval period. Sedge pollen is relatively lower at present (Figure 6.21).

One explanation could be the combined effect of wetland drainage and climate change. For earlier periods there is always a probability of finding sedge pollen on poorly drained terrain that is suitable for wetlands, despite the drier climate. Interpolation, based on scarce and non-uniform sedge pollen data, produced pollen surface artifacts in the northwestern corner of the state for the period 9000 B.P. An upward trend at the northwest corner of the state resulted in interpolated values exceeding 100% (Figure 6.21). Therefore, sedge pollen data may not serve as a reliable indicator of climate change.

Several questions are raised by close examination of the pollen surfaces. For example, oak pollen surfaces show higher concentrations of oak pollen in the northwestern corner of the state than in central Minnesota. Again, this may be attributable to the spatial irregularity in the distribution of pollen samples. For that reason, low pollen values for central Minnesota may reflect sparse data. Another source of discrepancy between pollen surfaces and climate trends may be that plant succession for temperate vegetation systems takes place in response not only to climate change, but also to periodic disasters or community dynamics (Ritchie 1981). There is also a lag between climate changes and vegetation response, which may partially confound interpretation.

Though generally successful, the pollen surfaces should not be taken literally. Factors controlling vegetation pattern at any given moment are so complex that the role of climate may be difficult to isolate. When paleoclimate reconstruction is implemented by reference to several kinds of evidence, reliability improves significantly. Several studies have demonstrated potential research directions, such as tracking past climate change by the integration of data on paleofauna, i.e. Quaternary mammalian sequences in alluvial valleys (Schultz and Martin 1970), or fish populations as an indicator of Pleistocene and Holocene environments (Cross 1970).

Pollen surfaces, interpolated from pollen core data, are generally consistent with Minnesota’s paleoenvironments as currently understood. Maps of the temporal - spatial dynamics of seven species depict known stages of climate and vegetation changes at a very coarse resolution. Of these, the best indicators of Minnesota’s paleoenvironments appear to be spruce, sage, and oak.

6.3.3.3 Palynological Variables

No variables were derived from these pollen surfaces because of the data problems described above and because the resolution, a function of the number and distribution of the core samples, was considered too low to contribute to the models. These data may be appropriate, however, for contributing to coarse resolution, statewide models that may be developed in the future.

Return to Top

6.3.4 Geomorphology and Geology

There were several sources of geomorphic data available at variable scales and with varying areal extents. The greatest level of detail came from the 1:24,000 scale maps developed for this project for seven major river valleys, one glacial lake bed and several upland quads in the state (Figure 6.23 and Chapter 12). These were not completed in time to be incorporated into the modeling. Landforms mapped for the Department of Natural Resources, at a scale of 1:100,000, were available only for the northern 2/3 of the state during Phase 2. In Phase 3, these became available for the entire state. However, additional interpretation was required before they could be used for deriving variables. The statewide bedrock, landform, geomorphology, and quaternary geology layers from MGC100 had a source resolution of 40 acres. They were suitable for regionalization and interpretation, for stratifying models, and for providing some general information on landscape features where higher resolution data were not available. Variables were derived from these layers for the Phase 2 and 3 models. Finally, paper and digital maps of bedrock geology at scales of 1:100,000 and 1:250,000 were used to determine locations of and calculate distances to culturally significant bedrock formations in the southeastern part of the state.

6.3.4.1 River Valley and Upland Landforms (landform/sediment assemblages)

Landforms of seven major river valleys, one lake basin, and six upland quads (Table 6.3) were mapped from 1:40,000 scale color infrared National Aerial Photography Program (NAPP) air photos onto 1:24,000 scale topographic maps (Figure 6.23 and Chapter 12). Landforms were mapped as landform-sediment assemblages (LfSA) to support analysis of their suitability for containing archaeological resources.

Table 6.3 River Valleys and Upland Landforms Mapped

RIVER VALLEY LANDFORMS (Mapped by river valleys)*	UPLAND LANDFORMS (Mapped by individual quad sheets)
Minnesota River*	Bemidji
St. Croix River (MN state line to mouth)*	Lake Agassiz
Rainy River (U.S. part only)*	Anoka Sand Plains
Red River*	Lake Benton
Root River*	Mountain Lake
Rock River*	Nicollet
Upper Mississippi River (Twin Cities northward)*
Red Lake (Big Bog) - mapped as a transect

In addition to the landform class of each unit, geomorphologists interpreted the potential of each unit for surficial and buried sites. Frequencies of site occurrence within specific landforms, calculated in the GIS, provided supplemental information for further contemplation of archaeological potential of landforms by the project geomorphologists. The landform classification scheme used for mapping river valley and upland landforms is outlined in Chapter 12. Because the final LfSA codes and suitability ratings were not available until the end of Phase 3 and these data were mapped for only a small portion of the state, this source could not be used to derive variables for modeling in Phases 2 or 3.

6.3.4.2 DNR Landforms

The Minnesota Department of Natural Resources contracted two entities for the mapping Minnesota's landforms at a scale of 1:100,000. Howard Mooers mapped the northern 2/3 of the state, and the Minnesota Geological Survey mapped the southern 1/3 of the state. MnDNR compiled the two sources to produce a statewide landforms coverage. These data became available only at the beginning of Phase 3. However, the attribute table provided was not suitable in its original form for deriving variables. The data were turned over to the project geomorphologists to establish relationships between their 1:24,000 LfSA classification system and records in the 1:100,000 DNR landforms coverage. Using the common code they developed, a new attribute table containing both the original DNR attributes and corresponding LfSA attributes was created at the end of Phase 3. The resulting database is suitable for deriving variables for future enhancements to Mn/Model.

6.3.4.3 Low Resolution Geomorphology (MGC100)

The geomorphic and geologic data in the MGC100 database were derived from the State Soil Atlas sheets at a scale of 1:125,000. These were converted to digital format with a resolution of 40 acres, and then resampled to a 100-meter cell size by the Land Management Information Center (LMIC). For this project, they were resampled again to a 30-meter cell size. The true resolution should be considered to be 40 acres. The geomorphology layers from this source include:

Geomorphic Regions: These are physiographic areas defined by topographic relief and soil parent material. There are 79 of these regions within the state. No variables were derived from this layer.

Landforms: This variable depicts types of geologic landforms represented by geomorphic regions. This is a generalization of the map of geomorphic regions. For example, the geomorphic regions Cass Drumlin Area, Pine River Drumlin Area, and Darling Drumlin Area are all classified as simply Drumlins in this layer. No variables were derived from this layer.

Quaternary Geology: This layer depicts unconsolidated sedimentary deposits of glacial or fluvial origin, which overlay the bedrock in much of the state. It allowed for the recognition of terraces, alluvium, colluvium, glacial lake sediment, and peat statewide. These features were used for deriving variables used in Phases 2 and 3.

6.3.4.4 Variables Derived from Quaternary Geology

Distance to Glacial Lake Sediment: Glacial lakes occupied much of Minnesota in the late glacial and early post-glacial periods. Fed by glacial meltwater, they eventually lost their sources and contracted. Glacial lake sediment and lake modified till represent areas that were formerly beneath glacial meltwater lakes. These lakes would have been sources of water in the early prehistoric landscape. Very early sites may be found near the edges of these ancient lakes. This Phase 2 and 3 variable measures the distance, in meters, to the edges of the sediment deposits, which presumably represent the maximum extent of the lakes themselves.

On Alluvium: Alluvium is material deposited by rivers along their floodplains as flood waters recede. The source layer mapped post-glacial (Holocene) alluvial deposits. Alluvium may bury archaeological sites. Alluvium may also contain artifacts that were transported by the river from their original locations. Presence/absence of alluvium was a Phase 2 and 3 variable with values of 1 for cells on alluvium and values of 0 for cells not on alluvium.

On Colluvium: Colluvium is material that has been deposited by gravity, for instance in a landslide or with slumping soil. The source layer mapped both glacial (Pleistocene) and post-glacial (Holocene) colluvium. Like alluvium, colluvium may either bury sites or contain artifacts that have been moved by colluvial processes from their original locations. Presence/absence of colluvium was a Phase 2 basic variable with values of 1 for cells on colluvium and values of 0 for cells not on colluvium. It was not used in Phase 3.

Glacial Lake Sediment: Presence/absence of glacial lake sediment was used for a Phase 2 variable on lake sediment with values of 1 for cells on sediment and values of 0 for cells not on sediment. This variable was not used in Phase 3. It was replaced with the variable distance to glacial lake sediment that measured distances to lake sediments in meters.

On Peat: Peat is partially decomposed vegetation, primarily mosses, rushes, and sedges that accumulate in cool wet places. Peat deposits typically occupy portions of marshes or shallow lake beds. They may be very deep (tens of meters) and cover large areas. The deposits mapped in this layer are of post-glacial (Holocene) origin and typically occupy the beds of former glacial lakes. The same cool wet conditions that prevent decomposition of the vegetation may also preserve archaeological artifacts made from wood or other organic materials. Presence/absence of peat was a Phase 2 basic variable with values of 1 for cells on peat and values of 0 for cells not on peat. This variable was not included in Phase 3 modeling.

On River Terraces: Terraces represent former river floodplains that are now at a higher level than the modern river. At the end of the last Ice Age, approximately 10,000 years ago, river channels carried large volumes of glacial meltwater. Those rivers were much wider and deeper than their present-day counterparts. Combining the declining levels of the glacial rivers with the down-cutting action of those rivers has resulted in abandoned floodplains that form terraces in modern river valleys. Because of the sequence of deposition and downcutting, higher terraces are older than younger terraces. Consequently, it would be expected that they would have a higher potential for very old sites than the younger, lower terraces. Presence/absence of terraces was used as a variable in Phases 2 and 3, with values of 1 for cells on terraces and values of 0 for cells not on terraces.

6.3.4.5 Bedrock Geology

Only the MGC100 database provided geologic data on a statewide basis. Of the available data layers, only depth to bedrock/outcrops was selected for modeling. This layer maps the depth to bedrock (in 100 foot increments) and the areas of significant outcrops from a 1:1,000,000 scale source. Some outcrops may have been sources of stone for tools.

Bedrock data were especially important for the southeastern portion of Minnesota, defined by the Blufflands and Rochester Plateau subsections. This area is unique in that it has not been glaciated since before the second to last glacial period (Illinoian). The terrain is rugged and deeply incised, with a number of rivers feeding into the Mississippi along the eastern border. This unglaciated, or driftless, area contains primary and secondary deposits of chert and galena that were important sources of raw materials for chipped stone tools. These quarry and workshop sites were not associated with water as were camp and village sites in the region.

Bedrock geology, mapped at scales of 1:100,000 and 1:250,000 were available for Southeastern Minnesota. Data were received in digital format from the Minnesota Geological Survey for Fillmore, Rice, and eastern Houston counties. These were created from source materials at a 1:100,000 scale. For the remainder of counties in the region, only the bedrock formations of interest were digitized from published maps in Phase 2. Dakota, Olmsted, and Winona Counties were digitized from 1:100,000 scale maps. Dodge, Goodhue, Houston, Mower, and Wabasha Counties were digitized from maps at a scale of 1:250,000. In addition, outcrop locations were digitized from published maps at a 1:100,000 scale for Dakota, Fillmore, Olmsted, Rice, and Winona counties. For the remaining counties, bedrock outcrops were derived from the MGC100 data. Outcrops were overlain on the bedrock maps to determine where outcrops of chert and galena occurred.

In Phase 3, bedrock data were further updated and refined to explicitly reflect information on the Galena Group (Prosser Limestone, Cummingsville and Stewartville formations) and the Prairie Du Chien Group (Shakopee and Oneota Dolomite formations), where it was assumed that outcrops of bedrock used for tools could occur. At this time, undifferentiated formations of both groups were incorporated into the coverages. In this Phase, it was assumed that outcrops would occur where these formations intersect with steep slopes and deeply cut river valleys. Areas of the above mentioned bedrock formations were overlain with steep slope and river valley grids to generate potential areas of bedrock outcrops.

6.3.4.6 Variables Derived from Bedrock Geology

Depth to Bedrock: This variable was taken directly from the MGC100 layer mapping the depth to bedrock (in 100 foot increments). This is a very coarse measure of depth. This statewide variable was used in Phase 2 only.

Distance to Bedrock Exposures: Outcrops for deriving this variable were derived by overlaying outcrops from the MGC100 and digitized layers with the digital maps of chert and galena formations. Distances measured to these outcrops were measured in meters. This variable was used only in the Phase 2 models for the Southeast Riverine East and Southeast Riverine West regions.

Distance to Bedrock Used for Tools: The source layer for this variable identified outcrops expected to occur where these chert and galena formations intersected with steep slopes and deeply cut river valleys. Distances to these exposures were measured in meters. The variable was used in Phase 3 modeling in four southeastern ecological subsections (St. Croix Moraines, The Blufflands, Oak Savanna, and Rochester Plateau). However, using the variable may not have been a wise choice outside of the Blufflands and Rochester Plateau, as high resolution bedrock maps were not available. Lower source data resolution resulted in very large variable values in northern peripheral parts of the modeled territory (Washington County). Some Phase 3 models identified an inverse relationship between sites and bedrock when high site probabilities were found in areas remote from bedrock used for tools. This relationship is suspect.

Return to Top

6.4 ARCHAEOLOGICAL VARIABLES

The archaeological database was the primary source of archaeological data for the project (see Chapter 5). Its records include both known archaeological sites from the SHPO files and random points from negative survey areas. In Phase 2, known sites from U.S. Forest Service files and random points generated by the GIS were added to the database. These were used for modeling, but were removed from the database before delivery to MnDOT.

The archaeological database is the source for information about locations of sites and nonsites, types of sites, sources of site location information, and quality of site information. It is also the source for the dependent variable (site presence or absence) in the model equation.

6.4.1 Site Presence

Information in this database permits the exclusion of some categories of sites from consideration in modeling. In this project, Phase 1, 2, and 3 models were built on samples that excluded single artifacts, which are considered to be more randomly distributed, therefore less predictable, than other types of sites. In addition, some Phase 2 models were built from samples that excluded both single artifacts and lithic scatters.

6.4.1.1 Role of Lithic Scatters

Background

A constraint of building predictive models of archaeological site location is the apparent random distribution of lithic scatters. Since precontact Native Americans in Minnesota traveled across all dry land areas of the state millions of small clusters of artifacts are distributed across the landscape. These most likely represent incidental and very short-term use of the land. Usually called lithic scatters, artifact scatters, or simply find spots, this category of site is usually defined as having very low densities of debitage and tools (e.g., less than 1 artifact per sq meter) and no culturally diagnostic artifacts. In this genre of study, they are called lithic scatters, because they consist for the most part of lithic (stone) artifacts, although fire-cracked rock, bone, and other culturally non-diagnostic material may be present. They can very widely in size, tool types present, type and variety of raw materials, and artifact density.

As lithic scatters become an increasingly expensive cultural resource management problem, they are becoming an ever more interesting research focus (Bintliff and Snodgrass 1988; Bradley 1987; Dunnell 1988; Dunnell and Simek 1995; Ebert and Camilli 1993; Frink 1984; Larralde 1988; Lewarch and O'Brien 1981; O'Brien and Lewarch 1981; Odell 1980; Schlanger and Orcutt 1986; Schofield 1991; Sullivan 1995; Wandsnider and Camilli 1992; Zvelebil et al. 1992). In contrast to the normal practice of assigning all lithic scatters to a functional taxon, such as limited-activity area or special-activity site, recent research has demonstrated that lithic scatters are most likely the byproducts of a wide range of activities, including acquisition of lithic materials (Green 1987), lithic reduction (Tainter 1979), procurement of edible resources (Goodyear 1977), processing of nonedible resources (Sullivan 1983), refuse disposal (Schiffer 1987), and agricultural activities (Wilkinson 1989). Archaeologists have attempted to reconstruct the formation pattern of individual artifact scatters (Odell 1980; Camilli 1988) and to integrate them into regional adaptive systems (Kelly 1988, Stoddart and Whitehead 1991; Sullivan 1995; Zvelebil et al. 1992). Other archaeologists have investigated the archaeological character of lithic scatters, the impact of repeated plowing on the structure of the scatters, and the reliability of surface collections, that is, the probability that the artifacts exposed on the surface and collected will be similar in successive collecting episodes (Shott 1995; Ammerman and Feldman 1978; Hasgrove et al. 1985). At the very least, these attempts to optimize the inferential potential of lithic scatters have demonstrated that, while they may not all contain 'significant' information, they cannot be ignored as a class.

A compromise position between ignoring lithic scatters and including them in 'site present' is to develop an explicit set of archaeological survey priorities for the statewide management of this class of site. The purpose of the priorities is "to increase the effectiveness of Phase I surveys in locating prehistoric sites which will contribute to our understanding of past cultural behavior" (Carr and Keller 1996). Whether and how this should be undertaken in Minnesota is a decision for the Minnesota SHPO, the Office of the State Archaeologist (OSA), MnDOT, and other governmental agencies. What is of interest here is the extent to which the inclusion of lithic scatter in the 'site present' category decreases the predictive power of the project's model.

Because most lithic scatters are plow-disturbed, non-stratified, multi-component precontact sites in upland areas, they are generally thought to "offer little if any new information to the understanding of past cultural behavior and would not be considered eligible to the National Register of Historic Places" (Carr and Keller 1996). Except for their contribution to settlement pattern studies, they are thought to produce no obviously significant date that would justify the predictive modeling perspective, since lithic scatters are located across the landscape, their inclusion in the category 'site present' weakens correlations between dependent and independent variables and results in a large amount of unexplained variability (Young et al. 1995). According to this argument, the exclusion of lithic scatters from predictive model databases will increase the predictive power of models and remove a site type that is insignificant and, therefore, does not require mitigation.

A difficulty in studying this problem is the definition of lithic scatter used in the Minnesota SHPO Archaeological Database and SHPO Lithic Scatter Context. In current SHPO usage, a lithic scatter is a site that (1) has no obvious surficial features; (2) lacks prehistoric ceramics; (3) contains no significant Euro-American artifacts; and (4) mainly contains stone flaking debris and tools, or both. They can contain bone and fire-cracked rock. A lithic scatter is distinct from a findspot (a single artifact) and an artifact scatter, which is any site with artifacts made from materials in addition to stone (e.g., ceramics). In this usage, a lithic scatter can be a very large site that contains culturally diagnostic material but has not yet been assigned to a functional category.

An alternative approach is to use area, density, and diversity measures. These were calculated for a small number of counties in an early stage of the Mn/Model project. These measures were eventually discarded because they are based on what is usually considered poor quality information. For example, site size, although recorded in the SHPO database, is often just an approximation, and artifact density is often a function of the intensity and frequency of surveys at a site. However, these are typical measures in a discipline based on uncertain knowledge (see Popper 1988 for a discussion of uncertain knowledge in science).

As used in this project, a 'lithic scatter' is simply a small site whose artifact content is not very dense or diverse (it should make no difference whether diagnostic artifacts, such as projectile points or pottery sherds, are present). A simple test of the random distribution hypothesis is to calculate the association of this class of site with the three zones of high, medium, and low probability for site presence. If 'lithic scatters' are 'everywhere,' they should be scattered across the three zones in the same proportions as those are found in the landscape. This would provide an argument for excluding them from at least the primary predictive models. Other simple measures of 'lithic scatter' could be adopted, too. For example, Young and her colleagues (1995) simply exclude all sites less than 5 square meters in area from their predictive models. Their size choice was based on comparative studies of small, single component hunter-gatherer sites, which show that they nearly always occupy an area greater than 5 square meters.

Lithic Scatters in Models

All models developed in Phase 2 and 3 excluded single artifacts (isolated finds) from the training data. Single artifacts are assumed to have a more or less random distribution, therefore they would be expected to contribute noise to the model. The case for lithic scatters, as they are represented in the Minnesota SHPO database, is less clear.

Since small lithic scatters as defined in the literature tend to be widely scattered across the landscape, it makes sense to exclude them from the modeled sites; including them usually masks site locational patterns of value for cultural resource management. In this case, however, problems exist in the definition of "lithic scatter" int he SHPO database that makes this category problematic. In the SHPO database, the term refers to both small lithic scatters, as the term is used in the literature, and any assemblage of lithic artifacts and/or debris whose function has not been determined. This raised the question of whether SHPO 'lithic scaters' should be included or excluded from the training data.

The argument for excluding lithic scatters is that, since lithic scatters are small lithic scatters as defined in the literature, these sites may simply contribute noise to the analysis and weaken the model. Many recent surveys, including those conducted for the MnSAS and Mn/Model, have located mostly small, limited activity sites. It is possible that the large, perhaps overwhelming, number of small lithic scatters and special activity sites may be masking the effects of some important variables for predicting more significant sites. It is the larger campsites that archaeologists may use to form their intuitive models of site location, simply because of their visibility and more restricted set of environmental characteristics. In many respects, these campsites also contain the highest information content of any sites. Some may argue that models built with lithic scatters in the database may be emphasizing the location of the smaller, perhaps less significant, sites at the expense of the larger ones. This problem will be addressed in future model evaluation and interpretation.

In Phase 2, models were run both with and without lithic scatters. Evaluation of the models indicate that including lithic scatters improves the gain statistic somewhat (Section 8.6.2 and Table 8.6.4). Lithic scatters were shown not to be randomly distributed with respect to Phase 3 site probability model classes, with 75% of all lithic scatters occurring in high probability areas. However, this indicator is biased, since lithic scatters were among the sites used to build the model in the first place.

6.4.1.2 Database Stratification

The archaeological database attributes were used to stratify the archaeological data for modeling. In Phases 1 and 2, known sites were divided into training and test populations based on the information available in the SHPO database. Sites located as the result of probabilistic or quality CRM surveys, or that had undergone a Phase 3 study, were assigned to the training database and used to build the Phase 1 and 2 models. All other known sites in the database were set aside and used to test the model. However, in Phase 3, the database was stratified randomly, rather than by data sources. Sites in the first randomly selected half of the database were assigned to the training data used in the first round of modeling while the second randomly selected half was reserved to test the model. In the second round of modeling, training and test datasets were switched to develop another set of models for comparison and further testing. In the third and final round of Phase 3 modeling, models were based on the entire archaeological database (excluding single artifacts), and no test population was present.

Return to Top

6.4.2 Site Absence

In predictive modeling, the dependent variable may simply be the presence or absence of archaeological sites or some quantifiable characteristic of sites. Site presence/absence is the simplest method and the one selected for this project. This method allows sites of different types and characteristics to be combined into one large database. Where the number of known sites is relatively small, particularly with respect to the range of environmental diversity, modeling presence/absence is a practical decision that may be necessary to achieve a large enough sample. With the presence/absence technique, known sites are assigned a value of 1 and nonsites are assigned a value of 0.

Nonsites can be represented in the models in one of two ways. In Phase 1, nonsites were represented by randomly placed negative survey points, located within areas that had been surveyed previously, but in which no site was found. The problem with this technique is that any bias in the location of surveys is transferred to this sample of points. The non-site locations would not be truly representative of the landscape as a whole and might even be similar to locations of sites. This assumes that archaeologists surveyed in places where they expected to find sites and not in other places. Therefore, the distinction between sites and nonsites might not be very great, producing relatively weak models. Evaluations of the Phase 1 models showed that this was the case in Minnesota. The distribution of model values among nonsites was often more similar to model values of sites than to the distribution of the model values throughout the region as a whole.

In Phases 2 and 3, truly random points generated by the GIS represented nonsites. These points reflect the landscape as a whole. Although they have not been surveyed, it can be assumed that only a very small number would contain archaeological sites, since site density is so low in our state (see Chapter 5). The Phase 2 and 3 models are stronger than the Phase 1 models, partly because the geographic differences between sites and nonsites are greater. However, models may still mostly reflect the kinds of places where archaeologists have looked for sites, rather than all kinds of places where sites may be. For this reason, it was important to develop a model of the types of landscapes not adequately surveyed to guide future survey decisions. The survey probability and survey implementation models were developed for this purpose in Phase 3.

Modeling selected characteristics of sites was also considered. For this approach, the characteristics must be meaningful, quantifiable and consistently recorded for a large number of sites. Artifact density and diversity are two measures that were suggested. Artifact density presumably would be a function of the length or intensity of occupancy of a site. Artifact diversity might indicate the variety of activities that occurred at the site. More significant sites would be assumed to have high values for density and diversity.

The primary problem with modeling density or diversity is in measurement. First, the concepts need to be defined clearly. Then experts must agree on criteria for their measurement. A classification scheme would be required, and this scheme must be consistently applied. Effective application requires having sufficient information about a very large number of recorded sites to determine the classification of each. Density and diversity were rejected for this project because they have not been adequately measured in Minnesota. The available archaeological database did not contain values for these characteristics. Moreover, existing records did not reliably contain information that could be converted to these measures.

A secondary problem with modeling density or diversity is the assumption that higher values are the equivalent of more significant archaeological sites. This assumption is flawed. For instance, Paleoindian sites are likely to have very low density and diversity values, but are still very significant because they are so rare. A single artifact that is a beautifully crafted, intact tool or pot might be significant as an object, but not as a site. Thus modeling on the basis of density or diversity may not yield the desired results.

6.5 Conclusion

Much of the modeling process consists of identifying and deriving variables from the available archaeological and environmental databases. Modeling success may be to a large degree a function of whether the appropriate variables have been selected for analysis and whether they were adequately derived. Although more than 70 variables were included in the Phase 2 modeling procedures, it became evident that this number should be reduced. The larger the number of variables, the more unwieldy the later modeling procedures become. Variables interact with one another in predictable and unpredictable ways. Almost every model required the elimination of several variables from the database before the statistical software would run (see Chapter 7). Superfluous variables also create 'noise' in the models. Variables may be related to site locations purely by chance, not because of any real relationship with hunter-gatherer subsistence, confounding model interpretation. Some of these problems were eliminated in Phase 3 by reducing the number of variables to 44. Further refinement of the variables could still be accomplished. This will be especially important as newly available data sources provide the potential for defining new variables.

Return to Top

REFERENCES

Ammerman, A.J., and M.W. Feldman
1978 Replicated Collection of Site Surfaces. American Antiquity 43:734-740.

Anfinson, S.F.
   1990 Archaeological Regions in Minnesota and the Woodland Period. In The Woodland Tradition in
       the Western Great Lakes: Papers Presented to Elden Johnson, edited by G. E. Gibbon, pp.
       135-166. University of Minnesota Publications in Anthropology No. 4. Department of
       Anthropology, University of Minnesota, Minneapolis.

Anfinson, S.F., and H.E. Wright, Jr.
   1990 Climatic Change and Culture in Prehistoric Minnesota. In The Woodland Tradition in the
       Western Great Lakes: Papers Presented to Elden Johnson, edited by Guy E. Gibbon, pp. 213-
       232. University of Minnesota Publications in Anthropology No. 4. Department of Anthropology,
       University of Minnesota, Minneapolis.

Bintliff, J., and A. Snodgrass
1988 Off-Site Pottery Distributions: A Regional and Interregional Perspective. Current Anthropology
19:506-513.

Borchert, J.R. and Gustafson, N.C.
1980 Atlas of Minnesota. Resources and Settlement. University of Minnesota, Publication No. CURA
80-5, 310 p.

Bourdo, E.A., Jr.
1956 A review of the General Land Office Survey and of its use in quantitative studies of former
forests. Ecology 37(4): 754-768.

Bradbury, J.P., and W.E. Dean (editors)
1993 Elk Lake, Minnesota: Evidence for Rapid Climate Change in the North-Central United
States. Special Paper 276, Geological Society of America, U. S. Geological Survey, Denver.

Bradley, R.
  1987 A Field Method for Investigating the Spatial Structure of Lithic Scatters. In Lithic Analysis and
       Later British Prehistory: Some Problems and Applications, Edited by A. Brown and M. Edmonds,
       pp. 39- 47. BAR British Series 162. B.A.R., Oxford.

Camilli, E.
1988 Interpreting Long-Term Land-Use Patterns from Archaeological Landscapes. American
Archaeology 7:57-66.

Carr, K.W., and J. Keller
  1996 The Development of Prehistoric Settlement Pattern Research Priorities in Pennsylvania. Unpublished
       manuscript. Bureau for Historic Preservation, Pennsylvania Historical and Museum Commission,
       Harrisburg.

Cowardin, L.M.
1977. Classification of Wetlands and Deep-Water Habitats of the United States (An Operational
Draft).

Cross, F.B.
   1970 Fishes as Indicators of Pleistocene and Recent Environments in the Central Great Plains, In
       Pleistocene and Recent Environments on the Central Great Plains, edited. by D. Wakefield
       and J.K. Jones, Special Publication 3, pp. 241-257, The University Press of Kansas, Lawrence.

Dunnell, R.C.
1988 Low-Density Archaeaological Records from Plowed Surfaces: Some Preliminary Considerations.
American Archaeology 7&:29-38.

Dunnell, R.C., and J. F. Simek
1995 Artifact Size and Plowzone Processes. Journal of Field Archaeology 22305-319.

Ebert, J.I., and E. Camilli
1993 Lithic Distributions and Their Analytical Potential: An Example. Lithich Technology 18(1&2):95-
105.

Farrand, W.R., Drexler, C.W.
   1985 Late Wisconsonian and Holocene History of the Lake Superior Basin. In Quaternary
       Evolution of the Great Lakes, edited by P.F. Karrow and P.E.Galkin, pp. 17-32. Geological
      Association of Canada Special Paper 30.

Frink, D.S.
1984 Artifact Behaviour within the Plow Zone. Journal of Field Archaeology 11:356-363.

Frison, G.C.
   1978 Prehistoric Hunters of the High Plains. Academic Press, New York. Governor’s Council on
       Geographic Information.
   1997 Developing Digital County Soil Surveys for Minnesota (It’s a Dirty Job). Prepared by the
       Soils Data Committee of the Minnesota Governor’s Council on Geographic Information. Minnesota
       Land Management Information Center, St Paul.

Goodyear, A.C., III
  1977 Helca II and III: An Interpretive Study of Archaeological Remains from the Lake Shore
       Project, Papago Reservation, South Central Arizona. Arizona State University Anthropological
       Research Paper No. 9. Department of Anthropology, Arizona State University, Tempe.

Green, M.
  1987 Chipped Stone Raw Materials and the Study of Interaction on Black Mesa, Arizona.
       Occasional Paper No. 11, Center for Archaeological Investigations, Southern Illinois University,
       Carbondale.

Hammer, J.
1993 A New Predictive Site Location Model for Interior New York State. Man in the Northeast
45:39-76.

Harrison, S.S., and J.P. Bluemle
1980 Flooding in the Grand Forks-East Grand Forks Area. North Dakota Geological Survey
Educational Series 12.

Haselgrove, C., M. Millettt, and I. Smith (editors)
1985 archaeology from the Ploughsoil: Studies in the Collection and Interpretation of Field Survey
Data. Department of Archaeology, University of Sheffield, Sheffield.

Hasenstab, R.
1991 Wetlands as a Critical Variable in Predictive Modeling of Prehistoric Site Locations: A Case
Study from the Passaic River Basin. Man in the Northeast. 42:39-61.

Heinselman, M.L.
   1974 Interpretation of Francis J. Marschner’s Map of the Original Vegetation of Minnesota. Published
       on the back of "The Original Vegetation of Minnesota" by F.J. Marschner. Compiled from U.S.
       General land Office Survey Notes. North Central Experiment Station, U.S. Department of
       Agriculture, St. Paul.

Kelly, R.L.
1988 Hunter-Gatherer Land-Use and Regional Geomorphology: Implications for Archaeological Survey.
American Archaeology 7:49-57.

Knox, J.C., et al.
1981 Holocene Fluvial Stratigraphy and Climatic Change in the Driftless Area, Wisconsin, In
Quaternary Paleoclimate, edited by W.C. Maheny, pp. 107-127. York University, England.

Larralde, S.
1988 The Timeless Survey: Problems in Defininf Component Assemblages. American Archaeology
7:8-12.

Lewarch, D.E., and M.J. O'Brien
  1981 The Expanding Role of Surface Assemblages in Archaeological Research. In Advances in
       Archaeological Method and Theory, edited by M. B. Schiffer, Vol. 4, pp.297-342. Academic
       Press, New York.

Marschner, F. J.
1974 The Original Vegetation of Minnesota. Compiled from U.S. General Land Office Survey
notes. North Central Forest Experiment Station, Forest Service, U.S. Department of Agriculture.

O'Brien, M.J.and D.E. Lewarch (editors)
1981 Plowzone Archaeology: Contribution to Theory and Technique. Vanderbuilt Publications in
Anthropology No. 27. Vanderbuilt University Press, Nashville.

Odell, G.H.
1980 Toward a More Behavioral Approach to Archaeological Lithic Concentrations. American
Antiquity 45:404-431.

Picha, P.R. and M.L. Gregg
1993 Chronostratigraphy of Upper James River Floodplain Sediments: Implications for Southeastern
North Dakota Archaeology. Geoarchaeology: An International Journal 8:203-215.

Ritchie, J.C.
1981 Problems of Interpretation of the Pollen Stratigraphy of Northwest North America, p. 377-391.
In Quaternary Paleoclimate, edited by W.C. Mahaney, pp. 377-391. York University, England.

Rychagov, G.I. (editor)
1984 World of Geography. Mysl, Moscow.

Sand, P., T. Eiber and A. Gansen
   1995 Native Trees of Minnesota: Using GIS to Compare Their Range to Ecological Factors.
       Presented at the Fifth Annual Conference of the Minnesota GIS/LIS Consortium, St. Louis Park,
       Minnesota.

Schiffer, M.B.
1987 Formation Proceses of the Archaeological Record. University of New Mexico Press, Albuquerque.

Schlanger, S.H., and J.D. Orcutt
1986 Site Surface Characteristics and Functional Inference. American Antiquity 51:296-312.

Schofield, A.J.
1991 Interpreting Artifact Scatters: An Introduction. In Interpreting Artifact Scatters: Contributions to
Ploughzone Archaeology, edited by A.J. Schofield, pp.3-8. Oxbow Books, Oxford.

Schultz, C.B., and L.D. Martin
   1970 Quaternary Mammalian Sequence in the Central Great Plains. In Pleistocene and Recent
       Environments on the Central Great Plains, edited by D. Wakefield and J.K. Jones, pp. 341-
       353. Special Publication 3, The University Press of Kansas, Lawrence.

Shott, M.J.
1995 Reliability of Archaeological Records on Cultivated Surfaces: A Michigan Case Study. Journal of
Field Archaeology 22:475-490.

Sibrava, V., D.Q. Bowen and G.M. Richmond (editors)
1986 Quaternary Glaciations in the Northern Hemisphere, In Quaternary Science Reviews, pp.
509-510, Pergamon Press, New York.

Stoddart, S.K.F., and N. Whitehead
  1991 Cleaning the Iguvine Stables: Site and Off-Site Analysis from a Central Mediterranean Perspective. In
       Interpreting Artifact Scatters: Contributions to Ploughzone Archaeology, edited by A.J. Schofield,
       pp.141-148. Oxbow, Books, Oxford.

Strahler, A.N.
1975 Physical Geography. Fourth Edition. John Wiley and Sons; New York.

Sullivan, A.P., III
  1983 Artifact Scatters and Subsistence Organization. Journal of FIeld Archaeology 22:49-64.
  1995 Storage, Nenedible Resource Processing, and the Interpretation of Sherd and Lithic Scatters in the
       Sonoran Desert Lowlands. Journal of FIeld Archaeology 10:309-323.

Tainter, J.A.
1979 The Mountainair Lithic Scatters: Settlement Patterns and Significance Evaluation of Low Density
Surface Sites. Journal of FIeld Archaeology 6:463-469.

Teller, J.T.
   1985 Glacial Lake Agassiz and its influence on the Great Lakes. In Quaternary Evolution of the
       Great Lakes, edited by P.F. Karrow and P.E. Galkin, pp. 1-16. Special Paper 30, Geological
       Association of Canada.

Trygg, J.W.
1964-1969 Composite Map of United States Land Surveyors Original Plats and Field Notes.
Published by Trygg Land Company, Ely, Minnesota.

University of Minnesota, Department of Soil Science, in cooperation with U.S. Soil Conservation
Service. Minnesota Soil Atlas. Map sheets at 1:250,000 scale. Agricultural Experiment Station,
University of Minnesota, St. Paul.

Wakefield, D.
   1970 Recurrent Climate Stress on Pleistocene and Recurrent Environment. In Pleistocene and
       Recent Environments on the Central Great Plains, edited by D. Wakefield and J.K. Jones.
       Special Publication 3, pp. 3-7. The University Press of Kansas, Lawrence.

Wandsnider, L., and E.L. Camilli
1992 The Character of Surface Archaeological Deposits and Its Influence on Survey Accuracy. Journal of
Field Archaeology 19:169-168.

Watson, P.J.
   1991 Origin of Food Production in Western Asia and Eastern North America. In: Quaternary
       Landscapes. Ed. L.C.K. Shane, E.J. Cushing, pp. 1- 37. University of Minnesota Press,
       Minneapolis.

Wilkinson, T.J.
1989 Extensive Sherd Scatters and Land-Use Intensity: Some Recent Results. Journal of Field
Archaeology 16:31-46.

Wood, W. R., and D.L. Johnson
   1978 A Survey of Disturbance Processes in Archaeological Site Formation. In Advances in
       Archaeological Method and Theory, vol. 1, edited by M.B. Schiffer, pp. 315-381. Academic
       Press, New York.

Wright, G.F.
1899 Man and the Glacial Period. D. Appleton and Co., New York.

Wright, H.E. Jr.
   1970 Vegetational History of the Central Plains. In Pleistocene and Recent Environments on the
       Central Great Plains. edited by D. Wakefield and J.K. Jones, Special Publication 3, pp. 157-
       172. The University Press of Kansas, Lawrence.

Young, P.M., M.R. Horne, C.D. Varley, P.J. Racher, and A.J. Clish
1995 A Biophysical Model for Prehistoric Archaeological SItes in Southern Ontario. Research and
Development Branch, Ministry of Transportation, Ontario.

Zvelebil, M., S.M. Green, and M.G. Macklin
  1992 Archaeological Landscapes, Lithic Scatters, and Human Behavior. In Space, Time, and
       Archaeological Landscales, edited by J. Rossignol and L. Wandsnider, pp.193-226. Plenum Press,
       New York.

Return to Top

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.