Skip common site navigation and headers
United States Environmental Protection Agency
Great Lakes Ecosystems
Begin Hierarchical Links EPA Home > Great Lakes EcosystemsUpland Ecosystems > 1994 Oak Savanna Conferences > Susan Will-Wolf
Aquatic Ecosystems
EPA Region 5 Critical Ecosystems
Ecosystem Funding
Great Lakes Basin Ecosystem
Great Lakes Biological Diversity
Green Landscaping
Rivers and Streams
Upland Ecosystem

1994 Proceedings
North American Conference on Savannas and Barrens


Susan Will-Wolf
and Timothy C. Montague
Department of Botany
430 Lincoln Drive
University of Wisconsin
Madison, WI 53706


Living in the Edge: 1994 Midwest Oak Savanna Conferences

Prior to European settlement, oak savanna and prairie covered about 50% of southern Wisconsin. Today only remnants remain. Ecologists and natural resource managers who undertake large-scale restoration of oak savanna landscapes need to know how savanna structure and species composition relate to environment, landscape, and disturbance. U.S. General Land Office (GLO) Survey notes compiled in the 1800's have been used to map presettlement vegetation, but they are also a useful source of information on vegetation-environment relationships (Leitner et al. 1991; and others).

We studied GLO survey notes (1833) for an area in south-central Wisconsin. The focus of our investigation was the identification of constraints by local environment and landscape characteristics on the distribution and structure of the witness trees, so the known shortcomings of the GLO surveys were of less concern to us than to others.

Links between environment and vegetation are best described as relational, not functional (Roberts 1989), so we discuss how environment constrains vegetation, rather than how it determines vegetation. Geomorphic processes constrain the physical environment which constrains the natural disturbance regime which constrains the vegetation. There are also vertical feedbacks between constraints; vegetation can constrain the disturbance regime and can also constrain itself.


We transcribed GLO survey records for six townships of northwestern Dane County, Wisconsin, and entered data on witness trees into a GIS database. Survey points were classified into ten types: eight wooded vegetation types based on witness tree combinations (pairs), and two non-woody vegetation types (marsh and prairie) where trees were absent. We calculated average tree size and density for each survey point.

Slope, aspect and elevation were read from topographic maps. A southness index was calculated from slope and aspect. Soil parent material, surface texture and water holding capacity for each point were taken from published soil surveys. Each point was designated glaciated or "driftless" (long-unglaciated); the study area straddles the boundary between the two regions. Local fire probability was based on fire breaks within one mile to the SW (170° to 270°) of each point. Regional fire probability was the standard deviation of elevations within three miles to the SW of each point. Fire intensity was the sum of potential fuel load (local soil and topography) and potential fire heat (southness and upwind firebreaks).

Variables were coded as categorical variables for analysis. Chi-square goodness of fit tests explored vegetation-environment associations. Chi-square tests of association were used to compare the environmental distributions of different vegetation types. Multiple logistic regression analyses explored multiple variable relationships.


The presettlement vegetation of the study area was a mosaic of upland oak forest (35%), savanna (52%), prairie (11%), marsh (2%) and a minor component (<1%) of non-oak forest and savanna. Bur oak was by far the most common vegetation type, 41% of the landscape, followed by white oak (19%), prairie (11%), and bur/white oak (10%). White oak points were 16% more common in the driftless area than in the glaciated region, while bur oak points were less common in the driftless area by almost the same amount.

Constraints On Vegetation Species Composition

Environment-Vegetation Comparisons. In general, vegetation groups were constrained by more environmental variables in the driftless area than in the glaciated region, and constraints in the driftless area were generally stronger. Differences between regions were stronger for bur oak and prairie than for white oak, whose distribution was strongly non-random with respect to most variables in both areas. There were several specific differences between the glaciated and driftless areas: 1) elevation was important in the driftless area but not in the glaciated area, 2) aspect and southness index were much more important in the driftless area, 3) fire intensity was important in both areas and 4) local fire probability was significant only in the glaciated area.

White oak avoided high nutrient, high water capacity sites on flat or southwest aspects, and high potential fire intensity in both regions (Fig 1a, c). In the glaciated region, white oak avoided sites with high local fire probability and favored glacial till. In the driftless region, white oak avoided alluvium and sites with high regional fire probability and favored high elevation sites. Bur oak points were associated only with high fire intensity in the glaciated area; in the driftless area bur oak favored low elevation sites with alluvium, flat to southwest aspects, and higher potential fire intensity (Fig. 1b,d). Prairie in the glaciated area was associated with high water capacity sites, nutrient-rich parent material, and sites with high potential fire frequency. Like other vegetation types, prairie points were more site-restricted in the driftless area.

Vegetation Similarity (Clique) Analysis. To further explore environmental constraints on vegetation composition, we adapted Roberts’ (in Allen & Hoekstra 1992) clique analysis. We plotted environmental specificity (chi-square p values) against vegetation homogeneity for less and less homogeneous groups of survey points (vegetation cliques) to explore how environmental factors compare in the degree to which they constrain vegetation. We started with homogeneous (single species) survey points, then added more and more dissimilar vegetation types (Fig. 1).

White oak (Fig. 1 a,c), for which most environmental factors acted as fine-grained constraints, contrasts strongly with bur oak (Fig. 1 b,d), which was only weakly constrained by environmental factors. The patterns seen for white oak are those expected where local environmental constraints are important. Bur oak was constrained much less by local environmental factors than by landscape fire regime, a signal clear even in the driftless area where fire regime was probably less dominating. In the glaciated region (Fig. 1a, b), local and regional fire probability constrained the broader sets of woody vegetation (all oaks, all trees) more than did the local soil or topographic factors. Thus, fire was a coarse-grained constraint on all woody vegetation there.

Differences Between Vegetation Groups. South's index was the single most consistently significant variable distinguishing vegetation types in both regions (Fig. 2). Soil parent material, water capacity, fire intensity and regional fire probability were also significant across the entire landscape. Topographic variables distinguished vegetation types more in the driftless region. Regional fire was the most consistently important fire variable, followed by fire intensity and local fire. Differences between glaciated and driftless regions were greatest for comparisons between prairie and treed sites.

White oak sites, as compared with bur oak sites, were found on steeper, more north-facing sites with poorer, drier soils and lower fire intensity in both regions. Soil surface texture was significant only in the glaciated region, and white oak sites were at higher elevations in the driftless region. White oak and bur oak were consistently different for the most variables of any comparison (Fig. 2).

White oak sites differed from prairie in having lower southness indices and regional fire probabilities across the entire study area. In the glaciated region, fewer white oak sites had fine soil surface texture. White oak and prairie showed greater divergence in the driftless region, where higher elevation, poorer soil parent material, lower water capacity, and lower fire intensity each distinguished white oak sites.

Bur oak and prairie had the fewest differences in their environments and were the only pair for which completely different variables were important in the two regions. In the glaciated region, prairie sites had higher local and regional fire probabilities, whereas in the driftless area they had richer, moister soils and more flat sites than bur oak.

Variable significance for prairie vs savanna (white + bur oak) was similar to white oak -prairie comparisons. Medium southness index and high regional fire probability consistently distinguished prairie sites across glacial regions.

Multiple Logistic Regression Analysis For Vegetation Groups. Logistic regression models tested for groups of environmental constraints that best distinguished vegetation types from all other points or which best distinguished pairs of vegetation types. Multiple models seldom included all variables which were significant alone (Fig 1 or 2), and sometimes included variables not significant singly. Alternate significant models were common, suggesting redundancy of information in the full ten variables.

Southness, elevation, soil water capacity, and the fire variables were the most frequently entered in models for both regions. Topographic and edaphic variables were slightly more prominent in driftless area models. Environmental constraints were quite different in the glaciated and driftless regions; not a single dependent variable had models with the same set of independent variables for both areas. Driftless area models had more variables than glaciated region models. White oak had more models and more variables per model than bur oak. Fire variables in general contributed to multiple regression models about equally for glaciated and driftless areas, a pattern not seen from single variable analysis. Local fire was important mostly in the glaciated area, while regional fire was significant in more driftless area models. Fire intensity contributed to models for both regions.

Constraints On Woodland Structure

Mean tree density at bur oak points was higher in the glaciated area, while mean density at white oak points was higher in the driftless area. Mean diameters of trees for both vegetation types were higher in the glaciated area than in the driftless area. Prairie, savanna and forest occupied relatively constant proportions of the landscape between glaciated and driftless areas. White oak points had larger trees in denser stands than expected in both regions, while bur oak had smaller trees than expected in both regions and lower density in the driftless region. Multiple regression models for tree size and tree density were generated for each species group in each region. Bur oak and white oak models included different predictor variables in each case. Fire variables were more important predictors for tree density than for tree size, appearing in all models. The species specificity of variable response patterns shown in the multiple logistic regression models means that discussion of general woodland structure patterns must be limited.


Like Curtis (1959), Leitner et al.. (1991), and others, we found that local site characteristics (slope, aspect, elevation and soil drainage) can be significant constraints on vegetation structure. We found that local site constraints were stronger in the driftless region, presumably where topographic diversity reduced the influence of fire. A major difference between our study and previous research on presettlement vegetation is our use of quantitative estimates of fire frequency and severity to detect differences in the fire regime of fire-prone vegetation types. Bur oak savanna and prairie occur on similar slopes, aspects and soils. Prairies, however, have significantly higher local and regional fire probabilities than bur oak savanna.

The strong contrast between white oak and bur oak was a major theme of our results. The apparent paradox of similar proportion of forest, savanna and prairie in the topographically heterogeneous driftless area and in the more homogeneous glaciated region, and the strong differences in species composition and structure plus evidence that variables constraining vegetation operated differently in the two landscapes are both linked to this contrast. Our results support the conventional wisdom that bur oak is more fire tolerant than white oak. Curtis (1959) and many others have noted that bur oak appears to be the most broadly tolerant of local environmental conditions of all the midwestern oak species. Our study of presettlement environmental constraints supports a stronger statement: that bur oak was adapted to a fire disturbance regime frequent and destructive enough to render constraint directly by local environmental variables inoperable as a mechanism for explaining distribution on the landscape. White oak, in contrast, showed distribution patterns and environmental constraints (Fig. 1) entirely consistent with a species constrained by moisture, nutrients, light availability and interactions with fire regime in competition with other tree species.


This research was funded by the Wisconsin Department of Natural Resources.


Allen, T. F. H. and Hoekstra, T. W. 1992. Toward a unified ecology. Columbia Univ. Press, New York, New York.

Curtis, J. T. 1959. The Vegetation of Wisconsin: An Ordination of Plant Communities. Univ. of Wisconsin Press, Madison.

Leitner, L.A., C.P. Dunn, G.R. Guntenspergen, F. Stearns and D.M. Sharpe. 1991. Effects of site, landscape features, and fire regime on vegetation patterns in presettlement southern Wisconsin. Landscape Ecology 5: 203-217.

Roberts, D. W. 1989. Fuzzy systems vegetation theory. Vegetation 83: 71-80.

Figure 1. Environmental specificity of vegetation groups with gradually decreasing homogeneity. Most homogeneous groups are to the left of each graph. A) White oak group to all vegetation in the glaciated region. B) Bur oak group to all vegetation in the glaciated region. C) and D) Equivalents for the driftless region.

Figure 1.  Environmental Specificity of Vegetation Groups with Gradually Decreasing Homogeneity


Figure 2. Single-variable comparisons of variable distributions for vegetation groups.

Figure 2.  Single-Variable Comparison of Variable Distributions for Vegetation Groups  


Begin Site Footer

EPA Home | Privacy and Security Notice | Contact Us