• bearing trees;
  • hierarchical structure;
  • landscape ecology;
  • spatial autocorrelation;
  • spatial scales;
  • species coexistences


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References
  • 1
    Bearing tree data were used to characterize the composition and spatial structure of the southern boreal presettlement forest in north-eastern Minnesota, United States of America. Data collected during the General Land Office Survey (GLO) between 1853 and 1917, represents 35 324 samples (each with 1–4 trees) in a 3.2 million-hectare landscape. Nine tree species contributed at least 1% to the overall composition. Individuals of white pine and red pine were larger than all other species and represented 9% of the tree population, while accounting for 27% of the standing basal area. Black spruce, paper birch and larch, the three most abundant species, collectively accounted for 51% of the population and 38% of the basal area.
  • 2
    Eight physiographic zones were characterized by differences in glacial histories, geological surfaces, soil complexes and topographic properties, and supported different compositional mixes of the nine important species.
  • 3
    Fifty-six per cent of all four-tree plots had three or four individuals of a single species. This level of conspecific aggregation is an order of magnitude greater than would be expected based on a random distribution of the same population of trees and species. Jack pine had the greatest plot-scale aggregation, with 45% of all plots containing jack pine trees having three or four jack pine individuals. Jaccard association of similarity values of species co-occurrence ranged from less than 0.05 to 0.24, indicating limited plot–scale interspecific associations.
  • 4
    Landscape spatial patterns of the species were measured at two spatial scales, 1–10 km and from 5 to 50 km. Conspecific autocorrelation patterns were positive while interspecific autocorrelation patterns tended to be either neutral or negative. Hence, plots dominated by any given species tended to spatially aggregate near other such plots, out to substantial distances.
  • 5
    Forest tree spatial patterns reflect complex fine to landscape-scale relationships involving environmental factors, disturbance events and regeneration strategies. Management and long-term conservation of forest landscapes should consider multiscale patterns in order to re-establish forest structural properties eliminated following the disruption of natural disturbance processes.


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

It is well known that every species has a unique suite of responses to environmental variation and that spatial distributions of species are thus individualistic by nature (Gleason 1926). Distribution patterns reflect processes that affect seed dispersal, safe sites for colonization, competitive interactions that emerge at both plot and landscape scales and sensitivity to disturbance events. Evaluation of hierarchical organization and the associated patterns of species segregation and co-occurrence (Urban et al. 1987) requires multiple scales of investigation because ecologically important associations, processes and functions observed at one scale may not be seen at others (Smith & Urban 1988; Wiens 1989; Podani et al. 1993; Wiens et al. 1993).

Species-poor boreal forests have not been extensively investigated beyond the plot-scale (but see Frelich & Reich 1995 and Kuuluvainen et al. 1998). Although, on first inspection forests appear to have a simple spatial structure, investigations of a Pinus sylvestris-dominated forest in Finland (Kuuluvainen et al. 1998) showed numerous complex fine-scale autocorrelated spatial patterns, and suggested that small-scale patterns would be expected to influence large-scale dynamic processes. Pacala & Deutschman (1995) simulated spatial distribution patterns in temperate forest, finding that the tree-scale spatial patterns could in theory influence system-scale biomass productivity. In this study, we examine composition and spatial patterns in the presettlement southern boreal forest of north-eastern Minnesota, United States of America (USA), to enable a better understanding of the initial conditions upon which human disturbance has subsequently acted and thus to provide a better framework for interpreting the causes and consequences of change in present-day forest conditions.

Colonization patterns of southern boreal species following stand-killing fires result from their variety of regeneration strategies. Some, such as balsam fir (Abies balsamea (L.) Mill), white pine (Pinus strobus L.), red pine (P. resinosa Ait), northern white cedar (Thuja occidentalis L.) and white spruce (Picea glauca (Moench) Voss) are dependent on seeds from the unburned perimeters, whereas jack pine (Pinus banksiana Lamb), aspen (Populus tremuloides Michx) and black spruce (Picea mariana (Mill) B.S.P.) burn and regenerate either from canopy-stored serotinous seeds or by vegetative means (Heinselman 1973, 1981). Among the conifers decreasing colonization is expected in the order jack pine, red pine, white pine, black spruce and balsam fir (Thomas & Wein 1985). Several site factors, including seed bed quality (Ahlgren & Ahlgren 1981), burn intensity (Ohmann & Grigal 1981) and competitive interaction for light and nutrient resources (Messier et al. 1999) influence the relative success of species following recolonization and a variety of spatial patterns therefore develop across fine to landscape scales (Burnett et al. 1998; Nichols et al. 1998).

Galipeau et al. (1997) attributed spatial structure in a Quebec-Ontario boreal forest to differences in identified colonization patterns between black spruce and balsam fir, due to seed dispersal, bimodal temporal colonization and delayed regeneration. Although density-dependent responses are often cited as the probable cause, site heterogeneity can also affect resulting spatial patterns. Houlé (1998), for example, examined the spatial-temporal autocorrelation trends of Betula alleghaniensis and showed that mapped seedling survival locations were independent of the preceding year’s seed rain.

One aspect of landscape ecology that has been largely overlooked is whether there are detectable landscape-scale spatial patterns in forests largely undisturbed by human activities. Undisturbed forest is restricted to small remnant patches and, even in boreal forests where human disturbance is minimal in some places, little is known about spatial patterns. Gustafson (1998) suggests that there are methodological limitations to quantification of presettlement spatial patterns. Associations of species with geological formations or soil properties, links to disturbance histories and influences of past land use patterns, have nevertheless been made using historical data such as General Land Office (GLO) records (Grimm 1984; Almendinger 1985; Iverson 1988), albeit not at the plot and landscape scale. Quantification of these patterns would, however, provide information on spatial-temporal dynamic patterns, interactions among species and multiple-species aggregation patterns. We can also test for the spatially-dependent correlated relationships across plot, region and landscape scales predicted by theoretical work (Andren 1994; Pacala & Deutschman 1995; Ives et al. 1998). We used GLO records, which although not ideal are the best presettlement forest records available, and have demonstrated value for landscape-scale vegetation reconstruction.

The presettlement forest of the Arrowhead Region in north-eastern Minnesota, USA (Fig. 1) was subjected to natural fire disturbance regimes. One of our objectives was to address the accuracy of the common preconception that because of the removal of white and red pine by logging, these two species had dominated the landscape (Larson 1949; Hoffman 1952; Flader 1983). Commercial logging started in the late 1800s and its ecology is therefore poorly documented. This paper evaluates composition, species co-occurrence and spatial distribution patterns in the presettlement forest to gain insight into its physiognomic characteristics. The following questions are addressed:


Figure 1. North-eastern Minnesota, USA. Arrowhead Region Physiographic Zones (after Wright 1972).

Download figure to PowerPoint

  • What was the composition and size structure of the presettlement forest?
  • How were the tree species associated with the physiographic zones and surface geological material?
  • How were the species associated with soil texture and permeability conditions?
  • Did conspecific and interspecific spatial patterns differ between the plot, intermediate and regional landscape scales?

Study area

  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

The 32 000 km2 Arrowhead Region of Minnesota (Fig. 1) is centred at approximately 575 910 East and 5 274 896 North UTM Zone 15. The region remains primarily forested and tree species of current major importance include aspen, pines, spruce, larch, balsam fir and paper birch.

Physiographic zones of Minnesota were mapped by Wright (1972) following precipitation, temperature and geological gradients which vary along both east–west and north–south transects (see Table 1 for characteristics of the study area). The region was glaciated until about 12 000 bp, and now contains numerous glacial moraines, glacially carved lakes, drumlins and exposed Canadian shield bedrock. Thin mineral soils derived from glacial tills and moraines are interspersed with Holocene peat and alluvial deposits. The area has a cold continental climate, with 30 years (1951–80) mean winter (December, January and February) and summer (June, July and August) temperatures recorded near the centre of the study area (Virginia, MN) of −13.4 °C and 17.5 °C, respectively (Baker et al. 1985). The ground is typically snow-covered from mid-November to mid-April. Historically, vegetation dynamics in this southern boreal forest were driven by fires initiated during heavy lightning storms in late summer (Heinselman 1973, 1996). Although Native Americans lived in this region, and species composition and spatial distribution patterns at the time of arrival of European settlers may reflect their use of resources, their influence is not considered here.

Table 1.  Dominant physiographic zones of north-eastern Minnesota, shown with the dominant geological materials and dominant soil properties. Numbers in ( ) are the proportion of the study area represented by the physiographic zone. Data are from Wright (1972)
Physiographic zoneDominant geological materialDominant soil properties
Border Lakes Region (26.0%)Rainy Lobe Vermilion Ground MoraineCoarse medium texture – low permeability
North Shore Highlands (14.5%)Superior Lobe Mille Lacs-Highland Ground MoraineMedium texture – moderate low permeability
Toimi Drumlin Area (11.8%)Rainy Lobe St Croix Ground MoraineMedium texture – moderate permeability
Aurora-Alborn Clay-till Area (4.9%)Des Moines Lobe Culver End MoraineHistisols – low permeability
Glacial Lakes Upham and Aitkin (5.6%)Holocene PeatsHistisols – low permeability
Chisholm Embarrass Area (10.6%)Rainy Lobe Vermilion Ground MoraineMedium texture – moderate permeability
Brainerd Automba Drumlin Area (4.9%)Superior Lobe OutwashCoarse medium texture – low permeability
Beltrami Arm Glacial Lake Agassiz (18.9%)Des Moines Lobe Erskine Moraine Lake Modified TillMedium fine texture – moderate low permeability


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

General Land Office records (Almendinger 1997) for the Arrowhead Region were integrated in a multipurpose geographical information system (GIS) database. Maps depicting the location of bearing trees, physiographic zones (Wright 1972), surficial geology (Morey et al. 1982) and soils (Cummins & Grigal 1981) are stored in ARC/INFO.

GLO records consist of spatially explicit information on tree species collected as part of the original Public Land Survey of the United States which mapped land holdings following the Northwest Ordinance of 1785. Each 6-mile (10 km) wide township-range unit, was subdivided into 1-mile (0.6 km) sections and then subdivided into quarter and quarter-quarter sections of land. Data were collected at the intersecting points within this nested mapping scheme, during a 64-year period prior to initiation of logging and local settlement in this area (1853–1917). Trees and/or cairns were used to mark the intersections. Specific instructions were issued by the Surveyors General of the United States (White 1983) and included requirements to mark four trees at each township section corner, and one in each adjoining section and two at each quarter section and meander corners, and to record species and size class data of each tree. Although standardized instructions were issued, the exact procedures varied with individual survey crews.

Although biases have been identified in GLO records (Bourdo 1956; Grimm 1981) and may constrain scientific analysis, they represent some of the best data available for investigating vegetation composition, distribution patterns, and species–environment and species–species association patterns prior to the large-scale land clearance by European settlers (Bourdo 1956; Grimm 1984; Delcourt & Delcourt 1996). The large sample size and landscape area considered, a small species pool and our conservative methodology, alleviate most of the potential bias in bearing tree data. However, statistical analysis, was not always possible, in which case exploratory data analysis techniques were used.

Forty species were reported in the bearing tree records for the north-eastern Minnesota presettlement forest. As in all GLO records, taxonomic ambiguities were plentiful, such as unreliable distinction between white and black spruce. However, we assume that then, as now, black spruce was far more common and that the GLO records of Picea generally refer to Picea mariana. Multiple species abbreviations were used for some species, as well as a single abbreviation for many others, and such errors were eliminated by combining entries into meaningful taxonomic units (Almendinger 1997).

Each GLO record was viewed as being analogous to a vegetation sample plot, although each GLO plot includes data for between only one and four trees. The database consists of 35 324 section and quarter section locations covering an area approaching 3.2 million hectares and spans across 365 townships. A total of 82 899 trees were recorded (at least one per plot), although surveyor instructions, incomplete surveys and incomplete data, as well as limitations encountered during the transcription of hand-written surveyor notes and conversion to digital formats, meant that only 26% had the maximum four trees. Most of our analyses consist of only these 9280 plots, yielding information for 37 120 trees.

We made a trade-off between spatial extent and spatial resolution. The study region was quite large, and fine-scale spatial data were not available for the entire study area. In all cases, the resolution associated with the bearing tree locations was substantially greater (fine-scale point-based digital maps) than the resolution of the environmental parameters (coarse-scale polygonal digital maps) with which we wished to link them. Map scales also varied with each of the environmental variables.

Count data were assembled for the nine most important species and the eight most important physiographic zones in the study region (i.e. those contributing more than 1% to the total area). A 9 × 8 matrix was cross-tabulated, using S-Plus (Statistical Sciences 1995) to give a goodness of fit statistic to determine whether species abundance patterns differed among the physiographic zones. Physiographic zones were also qualitatively examined for differences in abundance of soil orders, and area proportion of soils classified by texture and permeability classes.

GLO records included size class data (measured in inches) for each tree recorded. Although some authors have calculated basal area/unit area (Anderson & Anderson 1975) using tree diameter and estimates of distance from section corners, species mean distance from section corners is a known bias inherent to GLO records (Almendinger 1997). Estimation of species density would therefore be erroneous. Diameter measurements can, however, be used to calculate a composite species basal area estimate (Shiver & Borders 1996).

Soil distribution patterns were defined according to the ‘Soils and Land Surfaces of Minnesota 1980’ digital map developed by Cummins & Grigal (1981). Soil texture and permeability conditions, classified using a five-level ordinal ranking scheme following a soil moisture gradient, were superimposed on the GLO plots. Species associations with soil texture (coarse, coarse-medium, medium, medium-fine and organic) and permeability classes (high, high-moderate, moderate, moderate-low and low) were determined through goodness-of-fit tests. Adjusted residuals (Abrams & Ruffner 1995; Agresti 1996) from this analysis were used to assess whether species were positively, negatively or neutrally associated with soil characteristics. An inverse area-weighted correction function was used to construct histograms of the proportion of each species among the soil texture and permeability classes. We did not attempt to adjust the vegetation and soil association patterns for fine-scale variations known to occur in soil properties in the study area.

Estimates of species co-occurrence were developed from the four-tree plot data set and used to assess species associations and their distribution on the landscape. Jaccard association of similarity scores were used to determine the strength of species co-associations.

Non-metric multidimensional scaling (NMDS, PC-ORD Version 4.0, MjM Software, Gleneden Beach, Oregon, USA, McCune & Mefford 1999) was then used to assess which species were associated with other species. This technique, which benefits from being non-parametric and having relaxed assumptions (Sammon 1969; Fasham 1977; Kenkel & Orloci 1986; Green et al. 1997; McRae et al. 1998), was used to rank similarity scores and map species in ordination space. Estimates of species co-occurrence were transformed to percentages before the analysis. NMDS was run, using the autopilot (slow and thorough) mode, which automatically reruns the best solution (McCune, personal communication). Autopilot was configured to run 40 iterations with real data and 50 runs with randomized data. The scree plot revealed three significant dimensions in the ordination. The optimal graphical representation was revealed with two dimensions (Fig. 2) providing the most useful interpretation of the distribution of species at this scale.


Figure 2. Species associations are shown in a non-metric multidimensional ordination mapping. Co-occurrence rates of species at individual plots were used to develop this figure.

Download figure to PowerPoint

The 9280 four-tree plot records were used to determine: (a) the average number of individuals per species per plot; (b) the proportion of all plots where the species occurred; (c) whether species were locally dominant at the plot scale; and (d) whether species were distributed across the landscape evenly among plots. Plots that contained three or four individuals of a single species were considered representative of aggregation locations for that species. This is a conservative estimate of species spatial aggregation, because the distance to individual trees from section corner points was moderate, averaging 6.7 m for all species, and there was no indication that trees closer to the section corner were either skipped or not selected in favour of other trees. Binomial expectations for the percentage of plots with three or four individuals of each species were calculated following cluster sampling for species proportions (Hayek & Buzas 1997; Goldberg 1986).

Spatial autocorrelation was used to determine if species distributions patterns were aggregated at scales ranging from 1 km to 50 km. The analysis consists of measuring the distances from a plot to all other plots, allocating each pair to an a priori defined distance class, and determining whether the two plots have species in common. The analysis is iterated once for each distance class. Spatial autocorrelation with nominal data is reported using the standard normal deviate (SND), for which values between −1.96 and +1.96 represent the 95% confidence limits for a neutral pattern. Positive associations are present when the SND is above +1.96 and values less than −1.96 indicate over-dispersion. The method is useful in describing the spatial association of species patterns with high concentrations of a single species and for comparing plots among species on a landscape. In patchy landscapes, SND typically decreases as the distance class increases (Sokal & Oden 1978; Frelich et al. 1993).

Spatial autocorrelation conducted using two different spatial lags, representing an intermediate landscape-scale (1 km to 10 km with 1 km steps) and a regional landscape-scale (5 km to 50 km with 5 km steps), allowed us to determine: (a) if plots populated by four conspecifics were autocorrelated across the landscape; and (b) if plots with four trees of one species were spatially correlated with conspecific four-tree plots of other species. Positive autocorrelation would indicate, respectively, that species distribution patterns were spatially aggregated, thus providing a measure of how homogeneous the forest was relative to the distribution of a given species or that the forest was heterogeneous, possessing characteristics indicating a patterned mosaic landscape.

We also assessed autocorrelation between high aggregation plots (plots populated with four trees of a single species) and mixed two-species plots (the species in the high aggregation plots and one other, each species being represented twice). Spatial patterns were assessed from 1 km to 10 km distances by 1 km steps. The objective of this analysis was to determine if plots dominated by pairs of species were spatially related to other plots where each species was highly aggregated. The analysis was conducted using high-aggregation plots for each species in turn and the corresponding mixed two-species plots. Positive autocorrelation offers a refined measure of any mosaic spatial structure indicated in the earlier analyses.


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

Composition and basal area

Bearing tree records for north-eastern Minnesota identified nine species present in amounts greater than 1% of the total (Table 2). Spruce was the most abundant species, followed by paper birch (Betula papyrifera Marsh.), larch (Larix laricina Du Roi, K. Koch), aspen and balsam fir. Jack pine (7.8%), white pine (6.3%), white cedar (6.1%), and red pine (2.7%) were also present in significant amounts.

Table 2.  Composition, mean d.b.h. and basal area of the nine species present in the presettlement forest with greater than 1% of the total number of trees
SpeciesPercentage compositionMean d.b.h. (cm)Percentage of individual ≥ 50 cm d.b.h.Percentage of total basal area
Black spruce20.718.2 0.313.4
Paper birch15.121.0 1.414.0
Larch15.019.2 0.611.0
Aspen10.818.3 1.1 7.7
Balsam fir 9.417.1 0.1 5.2
Jack pine 7.819.0 0.6 5.7
White pine 6.340.027.620.1
Northern white cedar 6.122.4 1.7 6.0
Red pine 2.737.022.0 7.3
Other 6.1 9.6

This southern-boreal forest was composed of small and medium-sized trees but red pine and white pine were substantially larger than the other species in this region (Table 2). In the presettlement record, 22% of the red pine and 28% of the white pine were at least 50 cm in size measured at 1.4 m above the ground, whereas other species were represented predominantly by smaller trees. Although white pine represented only 6% of the trees, it accounted for 20% of the total basal area in the bearing tree records. Red pine produced 7% of the basal area while contributing only 3% of the composition. Collectively, the three most abundant species, black spruce, paper birch and larch (51% of all trees), accounted for 38% of the total basal area of standing timber over the entire landscape (Table 2).

Physiographic zones and surficial geology

The eight physiographic zones (Fig. 1) included in this analysis are characterized by differences in the abundance and basal area of bearing tree species, surficial geological materials, soil texture, soil permeability and climatic conditions, but were uniformly dominated by the same nine species (average of 95% of the composition in each zone). Each physiographic zone, however, supported a different mixture of the species and proportional amounts of basal area (Table 3), indicating that the presettlement forest composition and structure was heterogeneous. Variations can be seen for instance in abundance of larch, jack pine and balsam fir among the Glacial Lakes Upham and Aitkin, the Toimi Drumlin Area, North Shore Highlands and the Border Lakes Regions (Table 3) and particularly in percentage basal area contributions of red pine and white pine in each of the physiographic zones. Chi-square tests demonstrated that the species composition and basal area distribution patterns (χ2 = 191.6 and 26.9, respectively, with d.f. = 63) were significantly different among zones. Heterogeneous environmental characteristics including substrate properties (Table 1), climate and varying disturbance histories are reflected in the species distribution patterns. For example, larch was abundant (43.2%) and jack pine rare (0.2%) in Glacial Lakes Upham and Aitkin, an area dominated by histisols with low permeability.

Table 3.  Species percentage composition (first number in paired columns) and percentage basal area (second number in paired columns) of the presettlement tree species within the primary physiographic zones
SpeciesPhysiographic zones
Border Lakes RegionNorth Shore HighlandsToimi Drumlin AreaAurora Alborn Clay-till AreaGlacial Lakes Upham and AitkinChisholm Embarrass AreaBrainerd Automba Drumlin AreaBeltrami Arm Glacial Lake Agassiz
Jack pine17.513.4
Red pine 3.811.5 0.30.4
Aspen13.4 8.8
Paper birch17.214.519.319.716.616.416.717.19.213.816.714.917.515.59.47.4
Northern white cedar 3.43.910.
White pine
Balsam fir 8.84.618.
Spruce spp.20.715.318.014.029.418.923.414.022.514.419.010.710.85.522.616.6
Larch 9.17.8 9.17.316.715.224.619.143.231.914.912.423.513.214.712.3

Associations among species and soils

Soil texture and permeability are often correlated soil properties, but their correlation with the soil-map units was far from universal. However, each soil-mapping unit contains numerous fine-scale inclusions which are integrated into single discrete classes of soil texture and permeability.

Despite this unqualified environmental heterogeneity, statistically significant patterns of association among the species and the soil factors were identified through goodness-of-fit tests. Species abundance patterns varied among the soil texture and permeability classes, and positive, negative and neutral associations between the species and environmental conditions were detected (Tables 4 and 5, Fig. 3). Ten of the 25 potential soil-class combinations (five texture classes × five permeability classes) occurred in this landscape and each of these showed a distinct pattern of species abundance (Fig. 3). Three general species groups were identified from this analysis. Jack and red pine were both positively associated with the better drained, coarse-textured soils, in direct contrast to spruce and larch, which maintained a positive association for the organic soils found in hydric landscape settings. The remaining five species (aspen, balsam fir, paper birch, northern white cedar and white pine) were more variable, although they generally exhibited neutral or negative associations with low permeability, organic soils, as well as with highly permeable soils, and tended to be positive in at least some instances with intermediate soil texture and permeability classes.

Table 4.  Species percentage composition within soil texture classes. Positive (+), negative (−), and neutral associations (n) were assigned following a test for conditional independence, based on adjusted residual values exceeding ?3? indicating a lack of fit of the HO in that cell
 Soil texture class
Species (percentage of area)Coarse (6.9%)Coarse-medium (44.4%)Medium (26.3%)Medium-fine (12.3%)Organic (10.2%)
Jack pine11.6+12.0+ 2.8− 3.7− 1.5−
Red pine 4.0+ 3.6+ 2.3− 1.2− 0.9−
Aspen 9.6−12.5+ 6.6−20.5+ 7.6−
Paper birch15.8+18.2+19.9+ 9.3−10.4−
Northern white cedar 5.3− 5.8− 7.8+ 9.7+ 5.6−
White pine 7.5+ 6.2− 9.8+ 6.4− 2.9−
Balsam fir 8.7−11.6+10.0n11.5+ 4.1−
Spruce spp.21.4n20.1−22.9+22.2n29.4+
Table 5.  Species percentage composition within soil permeability classes. Conventions as in Table 4
 Soil permeability class
Species (percentage of area)High (2.8%)High-moderate (1.8%)Moderate (24.5%)Moderate-low (53.4%)Low (17.5%)
Jack pine10.8+11.9+11.0+ 0.6− 2.0−
Red pine 5.1+ 3.6+ 3.5+ 1.0− 0.6−
Aspen 8.5−10.1−14.0+ 9.0− 5.5−
Paper birch13.4−16.8n16.4n18.7+ 7.5−
Northern white cedar 6.2n 4.9− 6.1− 8.8+ 6.0n
White pine 5.2− 8.4+ 5.6n10.6+ 1.8−
Balsam fir 8.2− 9.010.2n12.8+ 4.1−
Spruce spp.23.5n20.6−21.5−20.8−30.9+

Figure 3. Histograms of species distributions patterns illustrating the response to soil texture and soil permeability conditions. An inverse area weighted corrected factor was used to eliminate the bias associated with unequal area of soil surfaces.

Download figure to PowerPoint

Species–species associations

Jaccard coefficients of interspecific association were calculated using the four-tree plot records (9280 plots) as a measure of the co-occurrence among species. Although a larger pool of 18 species was used to calculate these coefficients, the additional nine, which did not contribute at least 1% to the forest composition, all showed very low interspecific association. The results that follow pertain only to the nine most important species.

Four arbitrary levels of interspecific association strength were defined based upon the resulting range of coefficients: high (> 0.24), moderate (0.24–0.10), low (0.09–0.05) and very low (< 0.05). The nine species formed 32 interspecific pairings (out of 36 potential), none of which were in the high association category; 10 were moderate, 12 were low associations, and 10 were very low. Spruce and larch, the two species most strongly associated with the hydric and mesic soils, co-occurred more frequently than any other species pair (Table 6).

Table 6.  Species co-occurrence classes, based upon the Jaccard coefficient of similarity. No species pairings were found in the high association class (> 0.24). Species pairs are listed in decreasing rank order from highest to lowest association
Moderate association (0.24–0.10)Low association (0.9–0.05)Very low association (< 0.05)
Spruce – larchBalsam fir – white pineAspen – red pine
Balsam fir – paper birchBalsam fir – northern white cedarJack pine – larch
Spruce – paper birchAspen – spruceWhite pine – northern white cedar
Aspen – paper birchAspen – white pineAspen – northern white cedar
Spruce – balsam firSpruce – northern white cedarJack pine – balsam fir
Paper birch – white pineLarch – balsam firRed pine – balsam fir
Aspen – jack pinePaper birch – jack pineRed pine – spruce
Red pine – white pineSpruce – white pineJack pine – white pine
Paper birch – northern white cedarAspen – larchRed pine – larch
Aspen – balsam firSpruce – jack pineRed pine – northern white cedar
 Jack pine – red pine 
 Red pine – paper birch 

Species distribution patterns were influenced by both species–species associations and environmental conditions, as can be seen by comparing the NMDS ordination and individual species distribution maps. NMDS methodology ordinates species associations based upon plot-scale species co-occurrence similarity ranks (Fig. 2), and thus in this case represents a finer scale assessment of species distribution patterns than do the individual species distribution maps (Fig. 4). The relatively large distances between species in the ordination is due to the tendency for conspecifics to aggregate and for interspecific co-occurrence to be unusual (Table 6). The ordination was sensitive to an environmental gradient reflecting soil moisture conditions, so that sites where spruce and larch were abundant are located on the ordination close to one another and in the opposite corner from jack pine and red pine, while aspen, paper birch, balsam fir, white pine and northern white cedar, which occupy sites of more intermediate soil moisture conditions, are located in the central portion.


Figure 4. Distribution maps of Jack pine (a), white pine (b), white cedar (c), red pine (d), aspen (e), balsam fir (f), paper birch (g), black spruce (h) and larch (i). Only those plots with three and four individuals of the same species at the plot are shown.

Download figure to PowerPoint

Maps illustrating the distribution patterns of each of the individual species, using plots where the species was present at least three times, illustrate more generalized distribution patterns over the entire regional scale (Fig. 4). Although the species distributions patterns were generally overlapping, it is clear that each species was also associated with broader-scale environmental conditions. For example, jack pine (Fig. 4a) and red pine (Fig. 4d) were more abundant in the northern portion of the study area (Border Lakes Region) where soils are rocky and thin, whereas the greatest concentration of white pine (Fig. 4b) and balsam fir (Fig. 4f) occurred within the North Shore Highlands with intermediate soil texture and permeability and climatic conditions moderated by Lake Superior lake effects. Aspen (Fig. 4e), while distributed across the entire region, had a high concentration in the north-west on lacustrine soils along the southern shore of Lake Vermillion. Even the most abundant species, black spruce (Fig. 4h), and larch (Fig. 4i) show areas of high concentration related to specific environmental settings associated with the Toimi Drumlins and the Glacial Lakes Upham and Aitkin physiographic zones (Fig. 1).

Plot-scale aggregation

Using the 9280 plots with four individuals, we present several metrics describing the relative frequency, plot-scale and landscape-scale species aggregation patterns (Table 7). Two factors influence the proportion of all plots where a species was present (which we call frequency, column B), i.e. the total population of each species (out of the 37 120-tree pool, which we call total abundance) and their degree of plot aggregation. We arbitrarily define plot scale numerical dominance as an occurrence of three or four trees of the same species at a plot (columns C and D). Aggregation is defined (column F) as the ratio between the proportion of all plots where the species was dominant (column D) and the binomial expectation of the species being dominant at any plot (column E). Ratios exceeding 1 indicate clumped distributions whereas ratios less than 1 indicate over-dispersion.

Table 7.  Species frequency, dominance and aggregation patterns among the 9280 plots with four individuals. For all species, the importance of per plot clustering is substantially higher than expected by random chance. Numerical-dominance is defined as an occurrence of three or four trees of the same species at a plot (columns C and D). Aggregation is defined (column F) as the ratio between the number of all plots where the species was dominant (column D) and the binomial expectation of three or four individuals of the species being present at any plot (column E)
 Average number of individuals per plot where presentFrequency of plots where the species is presentPercentage of plots where numerically dominantBinomial expectation of 3 or 4 individuals of the species per plot for all plotsAggregation, i.e. ratio of observed numerical dominance: the level predicted by the binomial probability
   Plots where presentAll plots  
Jack pine2.413.3%45.1% 6.0%0.21% 28.6
Red pine1.9 6.0%26.2% 1.6%0.009%177.8
Aspen2.021.2%31.7% 6.7%0.53% 12.6
Paper birch1.834.0%21.9% 7.4%1.4%  5.2
Northern white cedar1.913.5%26.7% 3.6%0.11% 32.7
White pine1.715.0%19.7% 3.0%0.12% 25.0
Balsam fir1.721.7%18.3% 4.0%0.31% 12.9
Larch2.228.2%36.6%10.3%1.4%  7.5
Spruce spp.2.141.1%33.2%13.6%3.7%  3.7

If species were distributed randomly, each would be expected to be numerically dominant in a small fraction of all plots but that fraction would vary tremendously between species. Spruce for example (due solely to its high abundance) would be expected to dominate plots approximately 400 times more often than red pine (3.7% of all plots vs. 0.009%Table 7). However, all species tended to aggregate, i.e. to dominate plots far more frequently than expected randomly (Table 7, column F). Red pine tended to aggregate at the highest rate, approximately 180 times more often than would occur at random, with jack pine, white pine and northern white cedar also occurring in aggregations far more often (25–33 times) than expected. In contrast, spruce, paper birch and larch occurred in aggregations less than eight times as frequently as would be expected by chance (Table 7, column F). This strong tendency of all species to aggregate results in an impressively high proportion (56%) of all plots being dominated by a single species, given that a value of less than 8% would be expected if all distributions were random.

Using plot-scale frequency of occurrence as a measure of species distributions, the two most widely distributed species were spruce and paper birch, which were found in 41% and 34% of all plots, respectively. The distribution of each species was primarily driven by its total abundance but aggregation also is influential. Paper birch was thus more widely distributed than larch (34% vs. 28% of all plots), despite a similar total abundance (15%), because it tended to aggregate less frequently (Table 7, column F). Red pine was the least widely distributed of the major species (6% of all plots).

The species also differed in their tendencies to be numerically dominant in plots in which they occurred (Table 7, column C), as well as in relation to all plots (Table 7, column D). Jack pine had by far the highest relative numerical dominance of plots where it was found (45% of all plots with jack pine had three or four jack pine individuals). This measure of numerical dominance is sensitive to both the aggregation tendency and the overall abundance. Red pine was the most aggregated, but its population size was very low and thus it dominated 26% of the sites where it occurred, which is still a very high percentage considering its overall abundance. In essence, its very low numbers made it less likely to dominate sites where it occurred, despite high aggregation. The opposite situation existed for spruce; it was the most abundant and least aggregated, and yet still numerically dominated 33% of all plots where it was present. Birch, white pine and fir numerically dominated roughly one-fifth of the sites where they were present, due to a combination of intermediate frequency and aggregation rates.

Not surprisingly, spruce, as the most abundant species, numerically dominated 13.6% of all plots (Table 7, column D). The greater aggregation of larch than birch led to a wider distribution of the latter (see above) but also resulted in larch numerically dominating 10.3% of all plots compared with 7.4% for birch. Red pine numerically dominated 1.6% of all plots, with the remaining species numerically dominating between 3% and 7% of all plots.

Landscape-scale spatial patterns

Positive spatial association was consistently found for all conspecifics across the landscape when measured at either the intermediate (1–10 km) or regional landscape (5–50 km) scales (Fig. 5). This pattern was evident among the presettlement forest species until extremely large distances, often beyond 20 km, were encountered. The strongest positive conspecific spatial patterns were found for aspen, larch and jack pine at both scales (Fig. 5). Spruce, balsam fir and white pine also consistently exhibited significant positive intraspecific spatial associations on both scales. Red pine, paper birch and northern white cedar were found to have weaker positive intraspecific spatial association, although these were often significant.


Figure 5. Intraspecific spatial autocorrelation correlograms, pairing all monospecific four-tree plots of each species. Individual species data are from a subset of the 9280 four-tree plots that contain location information for a single species.

Download figure to PowerPoint

Examination of interspecific patterns (Fig. 6) revealed substantially less positive spatial autocorrelation and patterns of species spatial autocorrelations were not consistent across all distance classes. Most interspecific pairings were found to exhibit neutral or slightly negative spatial patterns across all distance classes. Three species pairs, however, have strikingly different patterns. Spruce–larch (Fig. 6a) are shown initially to be neutrally autocorrelated to 4000 m, yet beyond this distance (out to 25 km) were positively associated at both scales (Fig. 6b). Aspen–birch and aspen–balsam fir exhibited negative associations at distance classes greater than 2000 m (Fig. 6a and Fig. 6b).


Figure 6. Interspecific spatial autocorrelation correlograms, pairing monospecific four-tree plots of two species. Individual species data are from a subset of the 9280 four-tree plots that contain location information for a single species.

Download figure to PowerPoint

Species showing similar associations for soil texture and permeability site conditions (Fig. 3), such as jack pine and red pine on coarse soils, or aspen, paper birch, white pine and northern white cedar on intermediately moist soils, were found to have negative or neutral interspecific spatial autocorrelation. These results indicated that although species might have similar associations for environmental factors, different factors may be involved in the regulation of spatial distribution patterns within similar environment settings. An alternative interpretation is that the mono-dominant aggregations are themselves randomly mixed. Thus, the expected interspecific spatial association suggested strictly by common preferences for similar site conditions (Fig. 3) was not a reliable predictor of spatial association among the species.

Autocorrelation patterns between high aggregation plots (plots with four conspecifics) and mixed two-species plots were conducted to examine spatial dependency patterns across transition zones where both species occurred (Fig. 7). In all but a few cases, species combinations were found to have neutral association patterns at spatial lags from 1 to 10 km. Departure from this pattern is seen within a few species-pairs, for example spruce–balsam fir and jack pine–aspen (Fig. 7b,d), which exhibit positive association for one to three consecutive distance classes. Patterns were also asymmetric, in the sense that traversing from high aggregation plots of white pine to mixed species plots of white pine and paper birch was not the same pattern as going from the mixed species plots to high aggregation plots of birch. The paucity of positive autocorrelation associations indicates that smooth transitions between two species were not common in the presettlement forest. Rather areas with high concentrations of a single species were found to be independent of mixed-species zones.


Figure 7. Mixed species spatial autocorrelation correlograms, pairing monospecific four-tree plots with plots containing two species, each occurring twice.

Download figure to PowerPoint

As a result of differences among species in overall frequency, plot-scale aggregation and landscape-scale spatial dependence, we can classify the nine species into three rough groups: (a) species which aggregated markedly and had patchy landscape distribution patterns (jack, red and white pine, northern white cedar); (b) species with intermediate levels of plot aggregation coupled with intermediate landscape distributions (paper birch, aspen and balsam fir); and (c) species that had the lowest (but still high) degree of aggregation and broad unconstrained distribution patterns reflecting high abundance (spruce and larch).


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

Much of the natural history and ecology of the southern boreal forest of north-eastern Minnesota has been understood for some time (Ohmann & Ream 1971; Heinselman 1973, 1981, 1996; Ohmann & Grigal 1979), but this study has used a hierarchical approach (Allen & Starr 1982;Urban et al. 1987; Levin 1992) designed to improve this knowledge base.

The origin of the complex, spatially structured patterns is based on the regional geology, soils, disturbance regimes, species life histories and regeneration strategies. However, the results indicate that abiotic environmental heterogeneity in this low-diversity forest was sufficient to influence species distribution patterns at fine and landscape scales. Significant spatial relationships occurred at the finest plot-scale where species aggregation was substantial and these patterns potentially influenced the landscape-level spatial patterns through regeneration processes that extended across relatively large spatial domains.

Physiognomic properties

Although the presettlement forest had relatively few tree species, numerous compositional, abundance and spatial patterns existed that collectively made this forest a complex system. The observed distribution patterns of the species were most likely influenced by spatial heterogeneity in biogeophysical properties and disturbance histories of the physiographic zones.

Understanding why white pine and red pine were less abundant yet much larger than the other species (Table 3) requires a comparison of species growth characteristics and adaptations to fire. Both pines can live longer than most of the sympatric upland species by almost 100 years, and their maximum life expectancy, approximately 300–350 years, approaches the longest fire return interval in this region (Heinselman 1973; Elliott-Fisk 1988). Secondly, morphological characteristics of both white and red pine as mature trees are conducive to surviving fire of moderate to low intensity. These features include thick fire-resistant bark that provides protection to the cambium, limited amounts of volatile resins, and self-pruning (Fowells 1965). Heinselman (1973) used tree rings and fire scars to construct stand origin maps of the Boundary Waters Canoe Area Wilderness. In many places, multiple fire scars corresponded to different fires, indicating that trees of these species survived multiple fire events (Heinselman 1973). Individuals of these species may also have inhabited sites that were less likely to burn or were likely to burn less intensely during fires than several of the other species (Carlson, unpublished data).

Plot-scale reproductive strategies and aggregation

Environmental conditions favouring plant species coexistence include widespread seed dispersal, frequent small scale disturbance events, wide ecological niche and unrestricted access to suitable microsites (Schmida & Ellner 1984; Green 1989; McLaughlin & Roughgarden 1993; Lavorel et al. 1994). Some characteristics of the study area appear to favour co-occurrence, yet the degree to which it developed at the four-tree plot scale was low. Although the species have marked preferences, most can occupy any of the available environmental conditions present at the plot-scale. Since all of the requisite conditions for species co-occurrence exist, why was plot-scale aggregation among conspecifics more common that expected under random conditions?

Plot-scale species aggregation patterns suggest strong fine-scale clustering of favourable regeneration safe sites for every species, coincident with patchy seed availability. All of the upland species are prolific seed producers. Both vegetative and sexual regeneration are well represented in this suite of species; however, it is worth noting that the five most abundant species are capable of vegetative regeneration while the less abundant species are not. Effective seed rain dispersal distances are generally short, conforming to the negative exponential function (Lavorel et al. 1994; Cornett et al. 1997) that promotes conspecific aggregation. Additionally many of the species vegetatively regenerate prior to fires (aspen, balsam fir, larch, black spruce, northern white cedar, paper birch), and following fires (aspen and paper birch), even if the above ground portion of the tree is killed (Heinselman 1981; Elliott-Fisk 1988). Below-ground reproductive tissue that survives lethal temperatures may represent an analogue of advanced regeneration normally associated with gap located seedlings and saplings. Regeneration strategies therefore suggest that subsequent generations should occur in close proximity to mature trees. When regeneration occurs prior to fires, whether seed-based or vegetatively, new recruits would be likely to occur within spatially autocorrelated distances of the parent trees, generating aggregated spatial configurations. Following fires, similar tendencies could increase the likelihood of aggregated patterns as well, although different species are likely to regenerate.

Jack pine and black spruce provide contrasting examples of regeneration strategies that may promote aggregating tendencies. Jack pine is reproductively mature at 20–30 years (Benzie 1977) and in this region the fire return interval in jack pine stands was about 50 years (Heinselman 1973). Seed rain from serotinous cones begins quickly following fires, providing a competitive advantage over seed of other species having delayed dispersal strategies. The marked tendency for jack pine to occupy well-drained, coarse-textured soils and to reproduce from mature individuals following crown fire may explain its very high aggregation rate. While jack pine can not reproduce vegetatively, black spruce is capable of regenerating by layering prior to fires, is relatively shade tolerant, and thus may regenerate from seeds between disturbance events, and also has semiserotinous cones providing reproductive opportunity in varying habitats. These characteristics may help explain its much lower, but still positive, tendency to aggregate than jack pine.

Aggregation by balsam fir, northern white cedar and white pine is most likely to have developed due to slightly different processes. Following fire, these species are not capable of vegetative regeneration from roots or stumps. Burn patterns, natural fire breaks and species life spans could collectively allow small old growth patches of these species to form in refugia in various locations on the landscape. These species are also moderately to very shade-tolerant as seedlings, and cast deep shade as individual mature trees or within patches (Cornett et al. 1998; Machado 1999). Thus, these species are more likely to regenerate beneath or near themselves under long return intervals than the other species. Moreover, in those locations that were by chance missed by fire, species may have had greater opportunities to regenerate and influence the local spatial patterns of other species. All three species sometimes occurred in habitats that were less prone to high-intensity fires.

Intermediate to landscape-scale patterns

Positive conspecific and neutral interspecific spatial patterns characterized the landscape-scale configuration of the presettlement forest. Temporal and spatial autocorrelation of disturbance patches overlaid on the physiographic features of a landscape can theoretically explain aggregations (Moloney & Levin 1996) and may well have been at work in this case. Landscape heterogeneity, simplistically divided between suitable and unsuitable sites, can regulate successful vegetation establishment following disturbance. Aggregation of individual species throughout the landscape increases when the suitable sites occur in a contagious spatial distribution. In effect, once established in suitable sites, feedback processes may influence successive establishment, leading to greater aggregating and potential enhancements to sites that make them more suitable. Alternatively, when suitable sites are aggregated in only a small fraction of the landscape, conspecific aggregating may decrease if the effective colonization rate is reduced (Andren 1994; Ives et al. 1998). The proportion of seedling safe-sites varies temporally because disturbance processes restructure (both temporally and spatially) the proportion of these sites on the landscape.

There was an underlying mosaic composed of co-occurring species spatially distributed with neutral spatial association patterns within the landscape matrix, as indicated by the autocorrelograms (Figs 5, 6 and 7). The matrix was shown to be dominated by species that maintained strong conspecific positive spatial autocorrelation patterns over large distances (Fig. 5). Thus, we view spatial autocorrelation as a parallel and indirect measurement of habitat variation when assessed at the intermediate and landscape scale, rather than a method of patch delineation (McGarigal & Marks 1995).

All nine important species tended to occur in plots with conspecific individuals far more often than would occur at random, and more than half of all plots were dominated by a single species. Moreover, the likelihood of high concentrations of different species at fine scales was low. Over 1–10 km and larger scales, positive spatial autocorrelation by a given species was common. Why would this have occurred? As suggested earlier, the presettlement forest could be considered substantially more homogeneous at subregional scales than at either fine or regional scales. At the largest scale considered in this study, the Arrowhead Region is made up of zones that differ in geology and hydrology, which produces a regional patchy environment. Much of the Arrowhead Region also includes a very high degree of fine-scale heterogeneity, such as is due to close juxtaposition of wetland and upland microsites. Perhaps at subregional scales, such as physiographic provinces or smaller 10–100 km2 zones, the fine-scale patchwork repeats itself within a unit that is somewhat more homogeneous in terms of geology, topography and soils than the entire region. Such a zone would then tend to have a high abundance of a given species. The landscape mosaic in the presettlement forest was composed of numerous zones where each species was capable of dominating large but highly discontinuous expanses of habitat. Interspersed within these zones were areas dominated by other species, but at densities lower than necessary to form positive autocorrelated species assemblages.


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

It is apparent that this low diversity forest included complex spatially structured properties. We were able to define numerous examples of fine-scale and landscape-scale composition and spatial patterns, which were likely to be related to patchiness of environmental factors, disturbance (fire), and their relation to species regeneration strategies. Fire was spatially and temporally heterogeneous and controlled the vegetation dynamics in this landscape (Heinselman 1973), probably influencing species composition, abundance patterns, regeneration strategies, and thus the spatial structure across each of these scales.

At the plot scale, each species responds to local disturbances bounded by tolerances to disturbance form and intensity and regeneration potential. These processes link plot and landscape spatial scales by coupling regeneration strategies and dispersal mechanisms of the mature trees to the future generations. Several authors have recently shown that plot-scale spatial aggregation patterns can help explain landscape patterns (Green 1989; Andren 1994; Moloney & Levin 1996; Ives et al. 1998).

White pine was the most highly sought after species present in the presettlement forest. The benefits derived from this species have left a legacy for the economic development of Minnesota and throughout the Great Lakes region. However, this development had an associated cost, a substantially reduced population of the species on the landscape (White Pine Regeneration Strategies Work Group 1996). Currently public concern and agency leadership are leading the drive to re-establish white pine to its former status in Minnesota forests. White pine, in the area investigated in this report, was a significant component of the forest community because of its size and dominant growth form rather than its local abundance, which was patchily distributed. This historical framework should not be ignored in planning and managing for white pine in the future.

More generally, the main issues raised in this study are relevant to the development of forest management plans examined from local and regional perspectives. Effective conservation of forest biodiversity may be enhanced by consideration of the spatial structural characteristics of the presettlement forests, in addition to composition. Adherence to management policies targeting single species that disregard landscape-scale compositional, structural and functional diversity may fail to achieve long-term conservation objectives. Since spatial patterns are scale-dependent, planning should follow multiscale approaches addressing local and regional landscape-scale issues.

Landscape structure is dependent upon local, regional and global scale processes. Hierarchy theory that attempts to link scale, process and ecological response, provides a conceptual basis for the design and analysis of complex systems (O’Neill et al. 1986; Levin 1992). However, critically important multiple scale investigations remain rare due to the numerous difficulties associated with defining and explaining ecological processes and properties that develop across scale boundaries. For example, regional disturbance patterns simultaneously have an impact on forests at several scales (individual trees, patches and landscape-wide systems), resulting in small-scale gap dynamics, homogeneous but patchy forest age structure, and variation in the local to regional biodiversity patterns. Furthermore, understanding how organisms respond to scale-independent processes is limited initially by the spatial and temporal scale from which we observe processes and secondly because our interpretations and predictions at best are extrapolations often implying erroneous linear relationships. Conservation of dwindling natural resources, species population dynamics, community composition and structure and nutrient cycling are all potentially affected by scale-dependent ecological processes that develop at scales other than where we can observe them. New research focusing on disturbance ecology and landscape-scale spatial-temporal patterns is required to improve our understanding of this complex issue. Management of species, landscapes and ecosystems will benefit from the new information.


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References

We wish to thank the Wilderness Research Foundation of Ely, Minnesota, the National Science Foundation, the National Council for Air and Stream Improvement, and the McIntire-Stennis programs for financial assistance. We also appreciate the editorial assistance provided by David Gibson, Lindsay Haddon and several anonymous reviewers.


  1. Top of page
  2. Summary
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Conclusions
  9. Acknowledgements
  10. References
  • Abrams, M.D. & Ruffner, C.M. (1995) Physiographic analysis of witness-tree distribution (1765–1798) and present forest cover through north central Pennsylvania. Canadian Journal of Forest Research, 25, 659669.
  • Agresti, A. (1996) An Introduction to Categorical Data Analysis. John Wiley & Sons, Chichester, UK.
  • Ahlgren, C.E. & Ahlgren, I.F. (1981) Some effects of different forest litters on seed germination and growth. Canadian Journal of Forest Research, 11, 710714.
  • Allen, T.F.H. & Starr, T.B. (1982) Hierarchy: Perspectives for Ecological Complexity. The University of Chicago Press, Chicago.
  • Almendinger, J.C. (1985) The late-holocene development of jack pine forests on outwash plains, north central Minnesota. PhD Dissertation, University of Minnesota, St Paul, Minnesota.
  • Almendinger, J.C. (1997) Minnesota’s Bearing Tree Database. Natural Heritage Information System, Section of Ecological Services, Division of Fish and Wildlife, Minnesota Department of Natural Resources, St Paul, Minnesota.
  • Anderson, R.C. & Anderson, M.R. (1975) The presettlement vegetation of Williamson County, Illinois. Castanea, 40, 345365.
  • Andren, H. (1994) Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos, 71, 355366.
  • Baker, D.G., Kuehnast, E.L., Zandlo, J.A. (1985) Climate of Minnesota Part XV: Normal Temperatures (1951–80) and their Applications. AD-SB-2777–1985. Agricultural Experiment Station, University of Minnesota, St Paul, Minnesota.
  • Benzie, J.W. (1977) Manager’s Handbook for Jack Pine in the North Central States. Forest Service General Technical Rep NC-32. US Department of Agriculture, St Paul, Minnesota.
  • Bourdo, E.A. (1956) A review of the general land office survey and of its use in quantitative studies of former forests. Ecology, 37, 754768.
  • Burnett, R.M., August, P.V., Brown, H.J., Killingbeck, K.T. (1998) The influence of geomorphological heterogeneity on biodiversity. I. A patch-scale perspective. Conservation Biology, 12, 363370.
  • Cornett, M.W., Puettmann, K.J., Reich, P.B. (1998) Canopy type, forest floor, predation, and competition influence conifer seedling emergence and early survival in two Minnesota conifer-deciduous forests. Canadian Journal of Forest Research, 28, 196205.DOI: 10.1139/cjfr-28-2-196
  • Cornett, M., Reich, P., Puetmann, K. (1997) Canopy feedbacks and microtopography regulate conifer seedling distribution in two Minnesota conifer-deciduous forests. Ecoscience, 4, 353364.
  • Cummins, J.F. & Grigal, D.F. (1981) Legend to Map Soils and Land Surfaces of Minnesota. Soil Series No. 110. Department of Soil Science, University of Minnesota, St Paul, Minnesota.
  • Delcourt, H.R. & Delcourt, P.A. (1996) Presettlement landscape heterogeneity – evaluating grain of resolution using general land office survey data. Landscape Ecology, 11, 363381.
  • Elliott-Fisk, D.L. (1988) The boreal forest. North American Terrestrial Vegetation (eds M.G.Barbour & W.D.Billings), 3462. Cambridge University Press, Cambridge.
  • Fasham, M.J.R. (1977) A comparison of nonmetric multidimensional scaling, principal components and reciprocal averaging for the ordination of simulated coenoclines and coenoplanes. Ecology, 58, 551561.
  • Flader, S.L. (1983) The Great Lakes Forest: An Environmental and Social History. University of Minnesota Press, Minneapolis, Minnesota, in association with the Forest History Society, Santa Cruz, California.
  • Fowells, H.A. (1965) Silvics of Forest Trees of The United States. Agricultural Handbook no. 271. US Forest Service, US Department of Agriculture, Washington, D.C.
  • Frelich, L.E., Calcote, R.R., Davis, M.B. (1993) Patch formation and maintenance in an old-growth hemlock-hardwood forest. Ecology, 74, 513527.
  • Frelich, L.E. & Reich, P.B. (1995a) Neighborhood effects, disturbance and succession in forests of the western Great Lakes Region. Ecoscience, 2, 148158.
  • Galipeau, C., Kneeshaw, D., Bergeron, Y. (1997) White spruce and balsam fir colonization of a site in the southeastern boreal forest as observed 68 years after fire. Canadian Journal of Forest Research, 27, 139147.DOI: 10.1139/cjfr-27-2-139
  • Gleason, H.A. (1926) The individualistic concept of the plant association. Bulletin of the Torrey Botanical Club, 53, 120.
  • Goldberg, S. (1986) Probability: an Introduction. Dover Publications, New York.
  • Green, D.G. (1989) Simulated effects of fire, dispersal, and spatial pattern in competition within forest mosaics. Vegetatio, 82, 139153.
  • Green, P.T., O'Dowd, D.J., Lake, P.S. (1997) Control of seedling recruitment by land crabs in rain forest on a remote oceanic island. Ecology, 78, 24742486.
  • Grimm, E.C. (1981) An ecological and paleoecological study of the vegetation in the Big Woods region of Minnesota. PhD Dissertation, University of Minnesota, St Paul, Minnesota.
  • Grimm, E.C. (1984) Fire and other factors controlling the big woods vegetation of Minnesota in the mid-nineteenth century. Ecological Monographs, 54, 291311.
  • Gustafson, E.J. (1998) Quantifying landscape spatial pattern: what is the state of the art? Ecosystems, 1, 143156.DOI: 10.1007/s100219900011
  • Hayek, L.C. & Buzas, M.A. (1997) Surveying Natural Populations. Columbia University Press, New York.
  • Heinselman, M.L. (1973) Fire in the virgin forests of the Boundary Waters Canoe Area, Minnesota. Quaternary Research, 3, 329382.
  • Heinselman, M.L. (1981) Fire intensity and frequency as factors in the distribution and structure of northern ecosystems. Proceedings of the Conference. Fire Regimes and Ecosystem Properties (technical coordinators H.A.Mooney, T.M.Bonnickerson, N.L.Christensen, J.E.Lotan & W.A.Reiners). Gen. Technical Report WO-26. US Department of Agriculture, US Forest Service, Honolulu, Hawaii.
  • Heinselman, M.L. (1996) The Boundary Waters Wilderness Ecosystem. University of Minnesota Press, Minneapolis.
  • Hoffman, D. (1952) Paul Bunyan. Last of the Frontier Demigods. University of Nebraska Press, Lincoln, Nebraska.
  • Houlé, G. (1998) Seed dispersal and seedling recruitment of Betula alleghaniensis: spatial inconsistency in time. Ecology, 79, 807818.
  • Iverson, L.R. (1988) Land-use change in Illinois, USA: the influence of landscape attributes on current and historic land use. Landscape Ecology, 2, 4561.
  • Ives, A.R., Turner, M.G., Pearson, S.M. (1998) Local explanations of landscape patterns: can analytical approaches approximate simulation models of spatial processes? Ecosystems, 1, 3351.
  • Kenkel, N.C. & Orloci, L. (1986) Applying metric and nonmetric multidimensional scaling to ecological studies: some new results. Ecology, 67, 919928.
  • Kuuluvainen, T., Jarvinen, E., Kokkanen, T.J., Rouvien, S., Heikkien, K. (1998) Structural heterogeneity and spatial autocorrelation in a natural mature Pinus sylvestris dominated forest. Ecography, 21, 159174.
  • Larson, A. (1949) History of the White Pine Industry in Minnesota. Use and Abuse of America’s Natural Resources. Arno Press, New York.
  • Lavorel, S., O'Neill, R.V., Gardner, R.H. (1994) Spatio-temporal dispersal strategies and annual plant species coexistence in a structured landscape. Oikos, 71, 7588.
  • Levin, S.A. (1992) The problem of pattern and scale in ecology. Ecology, 73, 19431967.
  • Machado, J.L. (1999) The role of resource availability in the growth, distribution, survival and physiology of cold temperate tree species. PhD dissertation, University of Minnesota, St Paul, Minnesota.
  • McCune, B. & Mefford, M.J. (1999) PC-ORD Multivariate Analysis of Ecological Data, Version 4. MjM Software Design, Gleneden Beach, Oregon.
  • McGarigal, K. & Marks, B.J. (1995) Fragstats: Spatial Pattern Analysis Program for Quantifying Landscape Structure. General Technical Report PNW-GTR-351. U.S. Department of Agriculture, U.S. Forest Service, St Paul, Minnesota.
  • McLaughlin, J.F. & Roughgarden, J. (1993) Species interactions in space. Species Diversity in Ecological Communities Historical & Geographical Perspectives (eds R.Ricklefs & D.Schluter), 8998. The University of Chicago Press, Chicago.
  • McRae, G., Camp, D.K., Lyons, W.G., Dix, T.L. (1998) Relating benthic infaunal community structure to environmental variables in estuaries using nonmetric multidimensional scaling and similarity analysis. Environmental Monitoring and Assessment, 51, 233246.DOI: 10.1023/A:1005943504335
  • Messier, C., Doucet, R., Ruel, J.C., Claveau, Y., Kelly, C., Lechowicz, M.J. (1999) Functional ecology of advance regeneration in relation to light in boreal forests [Review]. Canadian Journal of Forest Research, 29, 812823.DOI: 10.1139/cjfr-29-6-812
  • Moloney, K.A. & Levin, S. (1996) The effects of disturbance architecture on landscape-level population dynamics. Ecology, 77, 375394.
  • Morey, G.B., Sims, P.K., Cannon, W.F., Mudrey, M.G., Southwick, D.L. (1982) Geologic Map of the Lake Superior Region Minnesota, Wisconsin, and Northern Michigan. Minnesota geological survey state map series S-13. Prepared in cooperation with the U.S. Geological Survey, Minnesota Geological Survey, St Paul, Minnesota.
  • Nichols, W.F., Killingbeck, K.T., August, P.V. (1998) The influence of geomorphological heterogeneity on biodiversity II: a landscape perspective. Conservation Biology, 12, 371379.
  • O'Neill, R.V., DeAngelis, D.L., Waide, J.B., Allen, T.F.H. (1986) A Hierarchical Concept of Ecosystems. Monographs in Population Biology 23. Princeton University Press, Princeton, New Jersey.
  • Ohmann, L.F. & Grigal, D.F. (1979) Early Regeneration and Nutrient Dynamics Following the 1971 Little Souix Forest Fire in Northeastern Minnesota. Monograph 21, Forest Science.
  • Ohmann, L.F. & Grigal, D.F. (1981) Contrasting vegetation response following two forest fires in northeastern Minnesota. The American Midland Naturalist, 106, 5464.
  • Ohmann, L.F. & Ream, R.R. (1971) Wilderness Ecology Plant Communities of the Boundary Waters Canoe Area. USDA Forest Service Research Paper, North Central Forest Experiment Station NC-63, US Department of Agriculture, US Forest Service.
  • Pacala, S.W. & Deutschman, D.H. (1995) Details that matter: the spatial distribution of individual trees maintains forest ecosystem function. Oikos, 74, 357365.
  • Podani, J., Czaran, T., Bartha, S. (1993) Pattern, area, and diversity: the importance of spatial scale in species assemblages. Abstracta Botanica, 17, 3751.
  • Sammon, John, W. (1969) A Nonlinear Mapping for Data Structure. IEEE. Transactions of Computers, C-18. 401–409.
  • Schmida, A. & Ellner, S. (1984) Coexistence of species with similar niches. Vegetatio, 58, 2955.
  • Shiver, B.D. & Borders, B.E. (1996) Sampling Techniques for Forest Resource Inventory. John Wiley & Sons, New York.
  • Smith, T.M. & Urban, D.L. (1988) Scale and resolution of forest structural pattern. Vegetatio, 74, 143150.
  • Sokal, R.R. & Oden, N.L. (1978) Spatial autocorrelation in biology, 1. Methodology. Biological Journal of the Linnean Society, 10, 199228.
  • Statistical Sciences (1995) S-Plus guide to statistical and mathematical analysis, Version 3.3. StatSci, A Division of MathSoft, Inc., Seattle.
  • Thomas, P.A. & Wein, R.W. (1985) The influence of shelter and the hypothetical effect of fire severity on the postfire establishment of conifers from seed. Canadian Journal of Forest Research, 15, 148155.
  • Urban, D.L., O'Neill, R.V., Shugart, H.H. (1987) Landscape ecology: a hierarchical perspective can help scientists understand spatial patterns. Bioscience, 37, 119127.
  • White Pine Regeneration Strategies Work Group (1996) Minnesota’s White Pine now and for the Future. A Report to the Minnesota Department of Natural Resources, Minnesota Forest Resources Council, St Paul, Minnesota.
  • White, C.A. (1983) A History of the Rectangular Survey System. U.S. Department of the Interior, Bureau of Land Management, U.S. Government Printing Office, Washington, D.C.
  • Wiens, J.A. (1989) Spatial scaling in ecology. Functional Ecology, 3, 385397.
  • Wiens, J.A., Stenseth, N.C., Van Horne, B., Ims, R.A. (1993) Ecological mechanisms and landscape ecology. Oikos, 66, 369380.
  • Wright, H.E. (1972) Physiography of Minnesota. Geology of Minnesota (eds P.K.Sims and G.B.Morey), pp. 561578. St Paul, Minnesota.

Received 19 March 2000 revision accepted 21 November 2000