We took four soil samples from each plot at 1.8 m from the plot center in the four cardinal directions to a depth of 8 cm, the maximum depth of fine roots measured in a subset of plots.

We estimated browsing to the nearest five percent as the proportion of browsed branch tips on shrubs. Grazing often removed all but the plant base precluding estimation of percent loss, so we scored grazing as a binary variable. Both browsing and grazing were estimated during the initial measurements of plant species cover.

We determined percent direct, indirect, and total irradiance for each sampling point from a hemispherical photograph using GLA software (1). We took photographs at the center of each plot at a height of 1 m and used this central photograph to represent the plot (3.99-m radius). We used a principal components analysis (PCA) to combine the components of irradiance into a composite estimate of light regime for the plot; the axis loadings for this PCA are given in Table 2.

We also performed a PCA on the 22 physiographic characteristics, and the first 5 PCA axes (explaining 64% of the variance in these characteristics) were included in our analyses as composite measures of physiographic variation among the plots. The weightings of these physiographic variables are given in Table 3, with a variable’s equilibrium weighting for any given axis calculated as Ö (1/22). This equilibrium weighting indicates whether a variable contributed significantly to a given axis (2). Slope positions are given at the micro- or mesoscale, with the former including a 40-m radius around the plot center (i.e., the local drainage pattern) and the latter pertaining to >200 m from the plot center (i.e., the topographic sequence at the scale of the reserve). In total, before removing insignificant variables, our environmental matrix consisted of 23 variables including soil moisture, 7 soil nutrients, soil pH, humus richness and depth, slope, browsing, grazing, north/south and east/west directions, as well as 5 PCA axes representing physiographic characteristics and 2 PCA axes representing light regime.

Canonical correspondence analysis (CCA) provides a number of statistics designed to explain the effects of the data on the CCA model and output, and also to show the relationship between the sample sites, species, and explanatory variables (3). Here we tabulate details of our CCA analysis that provide more information about the specific relationships between each species and the environmental variables used in the explanatory matrix. In particular, we present selected results from the global analysis with only environmental variables used in the explanatory matrix to demonstrate a simple interpretation of CCA results. Outputs are from CANOCO V.4.0 (3), with scaling based on interspecies distances to maintain species c ² distances (2). Species biplot scores in four dimensions are given in Table 4. These scores indicate the position of the species along each ordination axis. Table 4 also indicates the cumulative fit of each species with respect to each axis. To assess the environmental regime favored by a particular species, information from Table 4 is used in conjunction with information in Table 5, which indicates where each environmental measure falls on a given axis. For example, reading the proportion cumulative fit for each species in the column labeled "Cumulative fit, axis 1" (Table 4) will indicate how well that species is described by that axis. Note that because cumulative fit is tabulated, it is the change in cumulative fit between axes that is relevant on axes 2–4. For a species that is well described by a given axis, the relationship between the species and the measured environmental variables is assessed with reference to Tables 4 and 5. Environmental variables with more extreme scores (positive or negative) relative to other environmental variables on the same axis dominate placement of species on that axis. For example, the two ferns Onoclea sensibilis and Matteuccia struthiopteris are well represented on the first axis (20% and 23%, respectively) and the second axis (34% and 39%). These two ferns show some level of differentiation on the first axis (biplot scores of 1.3 and 1.8, respectively; Table 4), which is mainly dominated by a water gradient (from Table 5). Because the water gradient is positively correlated with the axis, this indicates that M. struthiopteris is found in wetter environments. These species also show very different scores on the second axis (–1.25 for O. sensibilis versus 1.26 for M. struthiopteris), indicating the association of O. sensibilis with more nitrate-poor sites.

The third-order polynomial that we used to model spatial patterns is commonly applied in trend-surface analysis (2). In this approach, site coordinates are first centered by subtracting the geographic centroid of all site coordinates, so that the center of the sampling map is assigned coordinates x = 0, y = 0. The centered Universal Transverse Mercator’s (UTM) of the sites are then used as predictor variables in a third-order polynomial regression model (i.e., species abundance = b₁x + b₂y + b₃x² + b₄xy + b₅y², etc.). Within the CCA, this approach will model unimodal or linear trends of species abundance along these variables. An example of a parabola-shaped pattern of species distribution modeled by the spatial component of the CCA is given in Fig. 5 for Solidago rugosa. This parabolic shape could be modeled with a quadratic or cubic trend in the x plane. The third-order polynomial model can describe spatial patterns that are not symmetric (i.e., not isotropic, but rather showing different changes in species abundance in different directions); such anisotropic spatial patterns are of interest because they are inconsistent with neutral theory.

For a full treatment of interpretation of ordination scores, see the CANOCO user manual (3) and ref. 2, which also treat the theory of the methodology. In addition, note that the scores representing the light regime are derived from a PCA as described in Methods, and that a more negative score indicates greater direct and indirect irradiance (i.e., the correlation in the PCA was negative).

Seedless vascular plant names are from ref. 4. All other nomenclature is from ref. 5.

1. Frazer, G. W., Canham, C. D. & Lertzman, K. P. (1999) GAP LIGHT ANALYZER (GLA) (Simon Fraser University, Burnaby, British Columbia, and the Institute of Ecosystem Studies, Millbrook, New York), Version 2.0.

2. Legendre, P. & Legendre, L. (1998) Numerical Ecology (Elsevier Science, Amsterdam), 2nd Ed.

3. ter Braak, C. J. F. & Smilauer, P. (1998) CANOCO (Centre for Biometry Wageningen, CPRO-DLO, Wageningen, The Netherlands), Version 4.0.

4. Flora of North America Editorial Committee, eds. (1993) Flora of North America North of Mexico (Oxford Univ. Press, New York), Vol. 2.

5. Gleason, H. A. & Cronquist, A. (1991) Manual of Vascular Plants of Northeastern United States and Adjacent Canada (New York Botanical Garden, New York), 2nd Ed.