David Gibson, Southern Illinois University Carbondale, Department of Plant Biology and Center for Ecology, Carbondale, IL 62901–6509, USA.
1Ecological phenomena exist at multiple scales, but measurements of diversity frequently consider only the smallest scale, that of the original sample plots. Additive partitioning of diversity allows multiple, hierarchical spatial scales of analysis to reveal the scale at which diversity is maximized.
2We examined the spatial partitioning of diversity across 378 permanent plots established in 10 7–369-ha research natural areas (RNA) in the 294 455-ha Shawnee National Forest, Illinois, USA. Diversity (richness and Shannon's and Simpson's indices) was partitioned across four spatial scales, i.e. within and between plots, and between RNA and natural divisions (corresponding to α and three levels of β diversity), for two strata of vegetation (trees and woody understorey).
3For both strata, the highest contribution to diversity measured as species richness was between plots and between RNA. Diversity was lower than expected within plots, although Simpson's and Shannon's indices achieved their maximum values at this scale. However, Shannon's index values were higher than Simpson's index values at the between-RNA scale and for all strata, indicating that the most common species were found at this scale.
4There was a simple asymptotic relationship between plot occupancy and local abundance, suggesting high colonization rates and rapid colonization of open habitat indicative of a wide niche breadth of the most abundant species.
5Synthesis and applications. The implications of these findings are that maximum diversity across a forested landscape is not necessarily at the scale of sampling (i.e. within plots) but may be at higher scales corresponding to larger landscape units. Moreover, the largest contribution of richness to total diversity occurred at a larger scale than diversity expressed using measures based upon information theory, which incorporate species abundance and evenness. Conservation efforts seeking to preserve diversity must identify and target the correct scales to allow effective management. In this study, management at and within the RNA represents the most appropriate scale for conserving maximum diversity.
In the temperate forest ecosystems of eastern North America, an understanding of the scaling of diversity is of particular importance as the effects of habitat loss and fragmentation need to be quantified (Gilliam, Turrill & Adams 1995; Keddy & Drummond 1996; Fuller 2001). Prior to European settlement, eastern North America was covered by primary deciduous forest (Braun 1950) but several factors have led to their decline. During settlement, most of the primary forests were cleared and much of what remains now is secondary forest (Cowell 1998). Because of the clearing of primary forest, suppression of fires and chestnut blight Endothia parasitica and Dutch-elm disease Ophiostoma ulmi, the oak–hickory (Quercus spp.–Carya spp.) forests of eastern North America are fragmented, compositionally altered and being invaded by other hardwood species such as sugar maple Acer saccharum and beech Fagus grandifolia (Bormann & Likens 1979; Fralish et al. 1991). Replacement of the dominant species impacts the entire ecosystem by changing the microclimate, understorey vegetation and fauna (Holling 2001). Fragmentation of forests impacts the ecosystem at local, regional and global scales (Fralish 1997).
Partitioning of total species diversity into additive components within and between communities provides a framework by which diversity can be measured at different levels of organization (Lande 1996; Godfray & Lawton 2001). The most frequently used diversity indices (i.e. Shannon's and Simpson's) are based upon information theory (Magurran 2004). These indices measure both the number of species (species richness) and the combined effect of richness and species relative abundance (Pielou 1969). These diversity indices can be used to compare patches and sites by linking spatial scales with diversity (Whittaker 1960, 1972, 1977; Willig 2001). Whittaker (1972), for example, quantified alpha (α), beta (β) and gamma (γ) diversity to describe diversity at different hierarchical scales, i.e. within plots, between plots and at the landscape level, respectively (Fig. 1). The total diversity in a pooled set of communities (γ diversity) can be partitioned into additive components within and between communities (Allan 1975; Lande 1996), which makes it possible to calculate the relative contribution of α, β and γ diversity across a range of spatial scales (Wagner, Wildi & Ewald 2000; Gering, Crist & Veech 2003).
Forest communities are dynamic, and understanding how diversity is partitioned at different scales will help in managing and maintaining the forests effectively across the landscape. Managers need an accurate measurement of diversity at all hierarchical scales. Partitioning of diversity helps researchers understand the scale or scales that are most critical for determining species composition and persistence. Identifying the scale or scales where maximum diversity occurs over time helps in understanding the vegetation dynamics (Small & McCarthy 2002) and will aid in planning forest management to conserve natural levels of diversity.
Our study was designed to partition diversity in a forested landscape at different scales. We used a randomization approach (Crist et al. 2003) to allow additive partitioning of diversity across a large regional landscape. Specifically, to address these issues of scaling and diversity, we partitioned diversity in 10 research natural areas (RNA) at different sampling scales in two natural, divisions of the Shawnee National Forest (SNF), Illinois, USA. Maintaining a constant unit of sampling (i.e. sample plots) while conducting analysis of plots grouped within and between sites and natural, divisions allowed the sample ‘grain’ (size of sample unit) to remain invariant while changing the sample ‘focus’ (area of inference) (Scheiner et al. 2000; Rahbek 2005). The sample ‘extent’, describing the geographical space over which comparisons were made in this study, was the landscape represented by SNF. The question addressed was how partitioning of species diversity at different focal scales (plots, sites and divisions) helps us understand diversity. In addition, we explored the relationship between local abundance and regional distribution across the SNF landscape to investigate further links between local and regional scales (Freckleton et al. 2005).
SNF (294 455 ha) in southern Illinois includes two natural divisions, the Ozark Hills Division (OZ, 50 000 ha, 39 807 in SNF) and the Shawnee Hills Division (SH, 243 306 ha in SNF), characterized by their glacial history and geology (Schuegman 1973). Ten RNA, ranging in size from 7 to 369 ha, were established between 1989 and 1991 across these two divisions as representatives of the communities in SNF. LaRue Pine Hills, Ozark Hill Prairies and Atwood Ridge RNA are part of OZ while Burke Branch, Cave Hill, Dennison Hollow, Panther Hollow, Barker's Bluff, Stone Face and Whoopie Cat RNA are part of SH (Fralish 1997). In OZ the highest altitudes are characterized by rugged topography and underlain by cherty limestone. The Ozark Hills are covered with up to 10 m of loess soil. Most of this area was logged between 1880 and 1920 (Marini 1995; Fralish 1997). In contrast, the Shawnee Hills are underlain with sandstone intermingled with limestone. This region was extensively logged throughout the 18–20th centuries with only 3–5% of the current forest being old growth (Parker & Ruffner 2004). Since logging ceased in the 1930–50s the forested areas have been largely undisturbed.
A total of 378 plots (the census data set), each with a radius of 11·4 or 17·8 m, was established from 1996 to 1998 in the 10 RNA. The RNA in OZ were sampled with 11·4-m radius plots as the topography was rugged and it was difficult to find large areas with uniform terrain. In contrast, six of the seven RNA in SH were sampled with 17·8-m radius plots. The terrain in Burke Branch RNA was rugged compared with the other RNA in SH and hence was sampled using 11·4-m radius plots. Plots were established using a stratified random design within each RNA and plot centres were permanently marked with a 1·5-m iron rod driven at least 50 cm into the soil. Plot locations were recorded on hand-drawn maps or, in many cases, with a global positioning system (GPS). The numbers of plots per RNA varied reflecting the size of the RNA. These plots were sampled and analysed initially at the RNA scale by Adams (1999), Grahame (1996), McCoy (1997), Shimp (1996) and Suchecki (1999).
Within each plot the diameter at breast height (d.b.h.) of each tree (woody individuals with d.b.h. ≥ 5·0 cm) was measured. These data were used to determine tree basal area and density in each plot. The density of the woody understorey (woody individuals > 2·5 cm and < 5 cm d.b.h.) was determined by counting all individuals in 5·6-m radius subplots for plots with a 11·4-m radius and 11·4-m radius subplots for plots with a 17·8-m radius in the centre of the permanent plot. The woody understorey included tree saplings, shrubs and woody vines. All nomenclature follows Mohlenbrock (1986).
Partitioning of diversity
We followed procedures outlined (summarized below) in Gering, Crist & Veech (2003) to partition diversity across spatial scales in our data. Separate analyses were conducted for the tree and woody understorey strata. The program partition (Gering & Crist 2002; Veech et al. 2002; Crist et al. 2003; Gering, Crist & Veech 2003; Summerville et al. 2003) was used to calculate diversity across the region, SNF. The observed α and β diversity was computed at each focal scale, where diversity was measured as species richness (No), Shannon's index (H′) or Simpson's index (λ). Shannon's H′ and Simpson's λ are presented as N1 = exp[H′] and N2 = 1/λ, respectively, to allow comparison with richness, where N1 approximates the number of abundant species and N2 approximates the number of very abundant species (Hill 1973; Peet 1974). Probabilities that the observed values for α and β diversity could have been obtained by chance alone were obtained by bootstrapping, allowing the statistical significance of the observed values to be tested. The density of all species from all samples at a given scale was combined to create a single species pool. Individuals were then randomly assigned to samples such that the initial density in the sample was maintained but resulting in a new number of taxa. The randomized samples were then partitioned to provide diversity measures at each scale. This randomization procedure was repeated 10 000 times to obtain null distributions of α and β estimates for the diversity measures at each scale of analysis. The observed values at the scale considered were compared against expected values generated from the null distribution obtained by the randomization procedure. The proportion of null values that were greater than or less than the actual observed estimates was used to assess statistical significance. The probabilities obtained from the randomization test were interpreted as P-values as in traditional parametric significance tests.
In this study, landscape-scale diversity across SNF is the sum of α and β diversity, where α is the average diversity within sampling units and β is average diversity between sampling units in the region, hence maintaining their traditional interpretation (Allan 1975; Wagner, Wildi & Ewald 2000). For example, α1 is the mean within-plot diversity whereas β1 is the diversity between plots and β2 is the diversity between RNA. Total diversity at the landscape level can be described as α1 + β1 + β2 + β3, i.e. average diversity within plots + diversity between plots + diversity between RNA + diversity between the two divisions (Fig. 1). Thus, the spatial focus varied from 0·04- to 0·1-ha plots, 7- to 217-ha sites (RNA), 39 807- and 254 648-ha divisions, up to the spatial extent of the 294 455-ha region of SNF (Fig. 1).
To partition diversity at various spatial scales, the individuals (trees or woody understorey) per species were counted for all of the 378 plots sampled across the region: 233 plots in OZ and 145 plots in SH. Diversity of the tree data was also partitioned in separate analyses of the plots in each division because the plots in the two divisions were of different sizes (11·4- or 17·8-m radius, respectively). The 39 plots from Burke Branch RNA were excluded from the separate analysis of SH as the plots in this RNA were of a smaller radius than those in the rest of the division. These separate analyses were conducted as a check to determine the effect of mixing two sample plot sizes on the partitioning of diversity.
For the woody understorey, two separate analyses were also conducted, one for OZ with plot sizes 5·6 m and one for SH with plot sizes of 11·4 m. The 80 plots at Atwood Ridge RNA were excluded from this separate analysis of OZ as the plots in this RNA were of a larger radius than those in the rest of the division.
The relationship between patch occupancy and local abundance in both strata was investigated by plotting (i) the proportion of sites occupied (P) against local abundance (N̂; basal area or density for trees and density woody understorey) and (ii) the relationship between local population abundance and regional population size, i.e. between N̂and Nτ = PN̂ (Freckleton et al. 2005). The nature of these relationships was investigated by fitting simple linear and non-linear regressions, with the latter being accepted when the variance explained (r2) represented an improvement of > 5% over the linear model.
Species accumulation curves
Species accumulation curves were constructed to evaluate the completeness of inventories in the two strata from the 378 plots with pc-ord (McCune & Mefford 1999). Average species accumulation curves were constructed by subsampling the entire community 500 times to determine the number of species as a function of subsample size. First-order jack-knife estimates of total species richness were calculated in PC-ORD as:
estimated total species richness = S + r1(n − 1)/n
where S is the number of species observed in n plots, and r1 is the number of species occurring in one sample plot (Palmer 1990). First-order estimates provide the highest levels of precision compared with other methods of estimating total species richness (Palmer 1991).
spatial scale patterns of diversity
There were 66 tree species in the entire landscape recorded from the 378 plots. Mean richness was significantly lower than expected (the pattern that would be found if individuals and species were distributed at random) within plots (Table 1 and Fig. 2a). The highest richness was observed between plots (β1 = 37%; Fig. 2a), although it too was significantly less than expected. Richness was significantly greater than expected between RNA (β2 = 31%) and between the two regions (β3 = 18%). In contrast, 80% of the diversity was partitioned within plots when expressed using Simpson's index, with small and decreasing amounts at increasing scales. When diversity was expressed using Shannon's index, the pattern of observed partitioning of diversity between scales was intermediate between that observed using richness and Simpson's index. Nevertheless, diversity decreased from the plot scale upwards as it did using Simpson's index. Diversity was, however, significantly less than expected by chance at the within-plot scale when expressed by either Shannon's or Simpson's indices (Table 1). There were only one to two (< 10%) abundant (N1, exp H′) and very abundant (N2, 1/λ) species (Hill 1973) observed at focal scales above that of the plots (Table 1). In contrast, within plots there were 5·5 and 4·17 abundant and very abundant taxa, respectively, representing 23–58% of the observed richness.
Table 1. Observed and expected additive partitioning of diversity of trees at four spatial scales for the full data set (census, n= 378 plots) and separate analyses of the Ozark Hills and Shawnee Hills divisions. In all cases pairs of observed and expected values are significantly different from each other (P < 0·0001)
Shannon's (exp H′)
Shawnee Hills (less Burke Branch)
The same patterns of partitioning of diversity were observed when the data from each division were considered in separate analyses to maintain constant plot size. Fifty-four and 41 tree species were recorded from OZ and SH, respectively. The mean richness per plot was 15·8 and 20·3 species, higher than expected (Table 1), i.e. the pattern that would be found if individuals and species were distributed at random. The highest richness was observed between plots (β1 = 57·4% and 39·5%) for both the divisions, as it was when the data from the two divisions were analysed together. In SH, observed diversity was significantly greater than expected between plots for all three measures, whereas in OZ observed diversity was less than expected at this scale when expressed as richness but greater than expected when expressed using Shannon's or Simpson's indices.
There were 76 woody understorey species in 367 plots (eleven plots of the 378 surveyed was dropped from the analysis as no woody understorey individuals were recorded) across the region. At the within and between plot (α1 and β1) scales, mean richness was 7·5 and 33·8, respectively, significantly lower than expected, indicating low diversity at these scales (Table 2). Between RNA and between divisions, the observed diversity was higher than expected, indicating greatest diversity at these scales. Partitioning of diversity between spatial scales was similar to that observed in the tree layer, with maximum richness between plots. When measured by Simpson's and Shannon's indices, the greatest diversity occurred at the plot scale and decreased with increasing scale (Table 2 and Fig. 2b). There was only 1.1–3.6 (< 10%) abundant (N1, exp H′) and one very abundant (N2, 1/λ) woody understorey species (Hill 1973) observed at focal scales above that of the plots (Table 2). In contrast, within plots (α1) there were 4·7 and 3·5 abundant and very abundant taxa, respectively, representing 47–63% of the observed richness.
Table 2. Observed and expected additive partitioning of diversity of the woody understorey at four spatial scales for the full data set (census, n= 367 plots) and separate analyses of the Ozark Hills and Shawnee Hills divisions. In all cases pairs of observed and expected values are significantly different from each other (P < 0·0001)
Shannon's (exp H′)
Ozark Hills (less Atwood Ridge)
When separate analyses for 152 plots from OZ and 95 plots from SH were analysed, within-plot (α1) mean richness was significantly lower than expected for both the divisions. Between-plot richness was also lower than expected for OZ and for SH. Richness was highest at the between-RNA (β2) scale (Table 2). When measured by Simpson's and Shannon's indices, the greatest diversity occurred at the between-plot and between-RNA scales for both divisions (Table 2). This pattern was similar to the analysis from the census data.
The most abundant species were also the most widespread across the region in both the tree and woody understorey layers (Fig. 3). The relationship between the proportion of sites occupied and local basal area and density was best fit with a non-linear regression in which occupancy showed a rapid initial increase but became saturating at high densities (Fig. 3). The relationship between regional population size (i.e. NT = PN̂) and local density (N̂), where P is the proportion of plots occupied (Freckleton et al. 2005), was linear (data not shown). Plots of species’ frequency distributions were unimodal, with most species being rare, with 40 tree and 61 woody understorey species occurring in less than 10% of the plots. Only a few species were widely distributed (see below).
In the tree layer, Quercus alba was both the most frequent species occurring in 69% of the plots and the most abundant (mean basal area 4·8 m2 ha−1, density 65 individuals ha−1). In addition to Q. alba, four other trees occurred in > 40% of the plots with a basal area > 1·0 m2 ha−1 (Carya glabra, Quercus rubra, Quercus velutina and Sassafras albidum). Cornus florida was widespread (63% occurrence) but not abundant (0·6 m2 ha−1), although its density was high (110 individuals ha−1). Pinus echinata and Magnolia acuminata were restricted to OZ, and Quercus prinus was present only on the dry upper slopes of both the divisions.
In the woody understorey, C. florida and S. albidum were the most widespread (42% and 38% of plots, respectively) and the most abundant species (678 and 419 individuals ha−1, respectively). Other widespread species in the woody understorey included Fagus grandifolia, Ostrya virginiana and Acer saccharum, occurring in 25%, 23% and 21% of the plots, with a density of 309, 247 and 181 individuals ha−1, respectively.
species accumulation curves
The species accumulation curves for both the tree and the woody understorey strata were starting to flatten towards a potential asymptote at our level of sampling intensity of 378 plots (Fig. 4). First-order jack-knife estimates of total species richness were 74 and 94 taxa in the tree and woody understorey strata, respectively. The observed richness of 66 and 76 tree and woody understorey taxa thus represented 89% and 81%, respectively, of the estimated total species richness.
We have identified the focal scales at which vegetation diversity occurs in SNF. This is important because it is argued that different processes determine diversity at different scales (Crawley & Harral 2001; Collins, Glenn & Briggs 2002). Our study is the first to partition diversity between different forest strata at a range of spatial scales (from 0·04-ha plots to a 294 000-ha landscape). The only comparable studies have been on the partitioning of insect diversity across two ecoregions in the mid-western USA (Gering, Crist & Veech 2003) and studies of agricultural fields in Switzerland (Wagner, Wildi & Ewald 2000) and Germany (Roschewitz et al. 2005). Gering, Crist & Veech's (2003) study indicated that the largest scale was most relevant for conservation and restoration of beetles. Wagner, Wildi & Ewald's (2000) study indicated that between-patch diversity contributed the most to overall plant diversity in an agricultural landscape, whereas Roschewitz et al. (2005) showed that arable weed diversity was maintained by high within- and between-field heterogeneity (β diversity). The results of our study are comparable to Wagner, Wildi & Ewald's (2000) in that we also found intermediate scales to have significantly higher than expected species richness and contain the largest proportion of diversity. Maximum richness was primarily between plots (β1) for both tree and woody understorey strata, followed secondarily by the between-RNA scale (β2; Tables 1 and 2). Bedrock defines the two divisions in SNF, which differ in species composition (Parker & Ruffner 2004), whereas the RNA were established to preserve both representative and, in some cases, unique features of the landscape. For example, LaRue Pine Hills RNA in OZ contains the only native stands of Pinus echinata in the region (Parker & Ruffner 2004).
Simpson's and Shannon's indices when expressed as a percentage were always highest at the plot scale, indicating maximum richness and evenness of species at this scale. Shannon's index was proportionally lower than Simpson's index at the plot scale, indicating that the most common species were not evenly distributed at this scale (Magurran 1988; Gering, Crist & Veech 2003). Simpson's index is the probability of drawing at random from a sample two individuals belonging to the same species (Gering, Crist & Veech 2003; Magurran 1988); as such it is more sensitive than Shannon's index to changes in the abundance of common species. In contrast, Shannon's index is sensitive to rare species (Peet 1974; Magurran 1988). Expressed as N1 (i.e. exp H′) and N2 (i.e. 1/λ), Shannon's and Simpson's indices reflect the number of relatively abundant and very abundant species, respectively (Hill 1973). Both these indices were higher at the α scale than the β scales because of local dominance and evenness of several species, including oaks and maples, giving them high information content at the plot scale. Crist et al. (2003) note that contrasting partitions of species richness and Shannon's index values reflect patterns of species dominance or rarity. Our observations are consistent with those of scale-related diversity relationships of forest Lepidoptera made by Summerville et al. (2003), suggesting that local factors determine species abundance at the within-plot scale whereas patterns of richness are determined at larger scales.
As in other systems and with other organisms (Gibson, Ely & Collins 1999; Collins, Glenn & Briggs 2002; Heino 2005), the most abundant species were also the most widespread in both strata (Fig. 3), indicating a link between local population processes and regional dynamics (Freckleton et al. 2005). The relatively few widespread, albeit abundant, species (e.g. Quercus alba and Cornus florida) and preponderance of infrequent species is consistent with the findings from partitioning diversity in which maximum richness was observed at the most local focal scale. The existence of a positive abundance–occupancy relationship provides generality for the findings presented here, and has implications for the species assemblages of the region (Gaston & Blackburn 2000). Several hypotheses have been proposed to account for abundance–occupancy relationships (Gaston & Blackburn 2000). The pattern observed for the forest species here (a positive abundance–occupancy relationship and a unimodal species frequency distribution) is most consistent with Brown's (1984) niche-based model predicting interspecific variation in realized niche breadths in which both regional distribution and local abundance reflect the degree to which local environmental conditions meet species’ requirements (Heino 2005). This pattern is consistent with models in which colonization rates are sufficiently high across the region so that all empty sites are immediately occupied (Freckleton et al. 2005). Indeed, the arrival rate of potential colonists to a site can directly affect scale-dependent relationships between local and regional diversity (Fukami 2004). When total diversity declines, the relative contribution of diversity at the smallest spatial scales to total diversity is maximized. The asymptotic relationship between mean local abundance and proportion of plots occupied implies a threshold above which increased local abundance does not allow increased site occupancy, consistent with a habitat-filling model (Freckleton et al. 2005).
Low richness at the smallest scales could be the result of a sampling effect, habitat fragmentation, low dispersal rates or microsite variation. Individual plots may not have been large enough to physically include all tree and woody understorey species (i.e. a sampling effect; Small & McCarthy 2002). However, this type of sampling effect is taken into account in the randomization part of the partitioning procedure. Expected richness of an individual plot cannot exceed the number of individuals in the plot, which is less than the total richness across the landscape (66 and 76 tree and woody understorey taxa, respectively, mean density of 49 ± 1·6 and 53 ± 2·7 trees and woody understorey individuals per plot, respectively). The fragmented nature of the eastern deciduous forest reduces regional colonization processes at the local scale and could reduce plot-scale richness (Fralish et al. 1991; Fralish 1997; Friedman, Reich & Frelich 2001). Although our study does not provide a test of this idea, our findings are consistent with models that postulate the operation of regional biogeographical processes governing levels of local diversity (Loreau 2000). Huston (1999) suggested competition, predation and environmental variability can reduce diversity at small scales, while mutualism and productivity can increase diversity. Low levels of local diversity, as we found for plot-scale richness, are considered in Hubbell's (2001) unified neutral model of diversity to be indicative of low rates of dispersal among species across the metacommunity represented by the landscape as a unit. A third factor that could contribute to low richness at the within- and between-plot scales (α and β1) is microsite variation between plots (Frelich, Machado & Reich 2003). At the larger scales all types of microsites were likely to be adequately represented, leading to high diversity (Beatty 2003). Studies that focus entirely at the plot scale are limited in the extent to which generalizations can be made to larger scales (Weiher & Howe 2003).
implications for conservation and management
Our study helps to advance the understanding of partitioning of diversity at different scales and strata across a large regional landscape by determining the focal scale at which the greatest contribution to total diversity occurs. This study unifies the disparate approaches to studying species diversity and composition, as all measures of diversity (especially β diversity) are expressed in the same unit. For all the strata, the highest richness was at the between-plot scale (β1), followed by the between-RNA scale (β2; Tables 1 and 2). The RNA were established to be ecological representatives of the larger landscape (USDA 1986) and this is well illustrated from our results and the partitioning of diversity. At the landscape scale, the species’ accumulation curves reflect adequate sampling as more than 80% richness in the tree and woody understorey strata were captured (Fig. 4). Although the RNA were established based on canopy cover and basal area of dominant trees, our results show that partitioning of diversity was the same for both strata, indicating that both are similarly structured ecologically in this region. Richness was lower than expected at the plot scale because the majority of diversity is at higher scales that cannot be captured in the smaller individual plots (Weiher & Howe 2003). There was an increasing disparity in the extent to which observed diversity exceeded expected diversity with increasing spatial scale (Tables 1 and 2). At the smallest scales, heterogeneity in distribution leads to exclusion of some taxa from local patches, resulting in observed diversity being significantly less than expected at the plot scale. With increasing scale there is an increasingly even distribution of taxa, allowing observed diversity to exceed expected. Thus, as scale increases, the apparent dominance of a few species that is seen at local scales becomes less apparent (Summerville et al. 2003).
Anthropogenic factors and suppression of natural disturbances, for example fire, have contributed to the fragmentation of the forest and composition change throughout North America (Cocke, Fulé & Crouse 2005). There is also increasing concern over the conversion of oak–hickory forests to sugar maple forests (Iverson 1994; Washburn & Arthur 2003). The relevant point in this study is that richness was highest at the between-plot scale, followed by richness at the between-RNA scale. The eastern deciduous forest occupies a vast area in North America and there is considerable variation in species composition and abundance (Braun 1950). For the preservation of the total diversity (γ diversity) in the landscape it is important to recognize and conserve diversity at the appropriate scales of interest (Keddy & Drummond 1996; Whittaker, Willis & Field 2001). By considering scales from the plot level to the landscape level this study has identified where diversity is maximized. When considering conservation or ecosystem restoration in this region, our results suggest that it is primarily within RNA that attention should be focused for conserving diversity. The RNA represent important conservation areas of the landscape and the high levels of diversity that we observed among sample plots allow managers and researchers to focus at a logistically tractable scale. However, some caution has to be exercised in making this interpretation because, as unique areas, the RNA may not be representative of the landscape as a whole. The RNA were initially identified on the basis of unique biotic features, including high species richness in some cases, and such areas, including many biodiversity hot spots (Latimer, Silander & Cowling 2005), merit high levels of protection. However, results such as those presented here reinforce the special nature of charismatic areas. Sampling in less high-profile areas of the landscape, outside the RNA in this case, would be necessary to assess the contribution of the background landscape matrix to regional diversity.
We thank the USDA Forest Service for partially funding the project, Eric Adams, Roger McCoy, Anthony Grahame, Jody Shimp and Paul Suchecki for collecting the data, Yohanes Honu and Allan Dzurny for their help in the field, and Loretta Battaglia, Kevin Davie, Tadashi Fukami, Yohanes Honu, Michael Hutchings, Peter Minchin, Charles Ruffner, Dale Vitt and Natalie West for comments on the manuscript. A special thanks to Joe Veech for the program partition, helping with the analysis and comments on the manuscript.