The roles of environmental filtering and colonization in the fine-scale spatial patterning of ground-layer plant communities in north temperate deciduous forests

Authors

  • Julia I. Burton,

    Corresponding author
    1. Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, 1630 Linden Dr., Madison, WI 53706, USA
    Search for more papers by this author
    • Present address: 321 Richardson Hall, Oregon State University, Corvallis, OR 97331, USA.

  • David J. Mladenoff,

    1. Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, 1630 Linden Dr., Madison, WI 53706, USA
    Search for more papers by this author
  • Murray K. Clayton,

    1. Department of Plant Pathology, University of Wisconsin-Madison, 1630 Linden Dr., Madison, WI 53706, USA
    2. Department of Statistics, University of Wisconsin-Madison, 1210 W. Dayton St., Madison, WI 53706, USA
    Search for more papers by this author
  • Jodi A. Forrester

    1. Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, 1630 Linden Dr., Madison, WI 53706, USA
    Search for more papers by this author

Correspondence author. E-mail: julia.burton@oregonstate.edu

Summary

1. The majority of plant species in northern temperate deciduous forests are restricted to the ground layer, but the importance of colonization processes relative to environmental filtering in structuring spatial variation in ground-layer plant communities is poorly understood.

2. Using multivariate analyses, structural equation modelling and geostatistics, we examined interactions among ground-layer plant communities, the live overstorey and environmental gradients across a 70- to 90-year-old northern hardwood forest in Wisconsin (USA). We hypothesized that (i) fine-scale variation is related to environmental filtering rather than dispersal limitation and colonization processes; and (ii) exogenous ‘site’ filters exert more control than the composition and structure of the overstorey.

3. A transition from communities of spring ephemerals to communities of evergreen-dimorphic species is related to a hierarchy of controls driven by elevation, soil texture and associated effects on soil moisture, overstorey composition, and O-horizon and soil properties. An orthogonal axis distinguished among sparse communities associated with high levels of soil moisture early in the growing season and rich communities of early summer forbs associated with increasing O-horizon N:P and %Ca, and short-distance dispersal mechanisms. Indirect effects of tree species are significant, but cumulatively less important than exogenous site filters.

4.Synthesis. The spatial patterning of ground-layer plant communities is related to both environmental filtering and colonization. These patterns were related to species’ functional and dispersal characteristics, suggesting that processes structuring ground-layer plant communities are not merely neutral. Loose regulation of environmental and resource gradients resulting in a coarse-grained spatial patterning of plant communities observed in second-growth forests may therefore be related to a simplification in overstorey composition and the absence of heterogeneity accumulating through gap dynamics.

Introduction

Understanding the relative importance of historical and environmental filters that drive variation in plant communities is a central goal for ecologists (Ricklefs 1987; Keddy 1992; Lambers, Chapin & Pons 1998; Harrison et al. 2006). Physiological filters can yield strong relationships between species distributions and environmental gradients, while biological filters (e.g. competition or predation), may constrain such distributions (McGill et al. 2006). In contrast, historical filters, such as disturbance or land-use history, can weaken relationships between species distributions and environmental gradients (Vellend et al. 2007) resulting in spatial patterns associated with colonization (Matlack 1994; Verheyen et al. 2003; Ozinga et al. 2005; Damschen et al. 2008) and ecological drift associated with stochastic processes (Hubbell 2001; Gotelli & McGill 2006). Studies of fine-scale spatial variation in plant community structure can provide insights into how such processes interact across scales to influence vegetation locally (Leibold et al. 2004).

Ground-layer plant communities of herb and shrub species comprise an important component of temperate deciduous forest ecosystems (Gilliam 2007), but are studied less frequently and intensively compared to tree species. Therefore, our knowledge of the physiology of individual species, and of species interactions and dynamics along environmental gradients, as well as our ability to predict the effects of changes in the structure of environmental gradients (i.e. due to disturbance and forest stand dynamics as well as climate change) is comparatively limited (Whigham 2004). Within temperate deciduous forests, ground-layer plant community structure has been associated with land-use history (Fraterrigo, Turner & Pearson 2006; Vellend et al. 2007), overstorey composition and structure (Beatty 1984; Scheller & Mladenoff 2002), disturbance (Moore & Vankat 1986; Roberts 2007), competition with tree seedlings and saplings (Miller, Mladenoff & Clayton 2002), interactions between earthworms and ungulate herbivory (Frelich et al. 2006) and soil properties (Rogers 1982; Beatty 1984; Reed et al. 1993), as well as colonization processes and seed dispersal mechanisms (Ehrlén & Eriksson 2000; Miller, Mladenoff & Clayton 2002). However, a synthetic model of the myriad of direct and indirect interactions that structure ground-layer plant communities has yet to be tested.

The relative importance of different environmental and dispersal filters for ground-layer plant community assembly may vary over space and time with land use and among forest stands in different stages of development. For instance, short-distance dispersal can constrain the ability of many forest herb species to re-colonize recovering stands in earlier stages of development (Matlack 1994; Verheyen et al. 2003). However, endogenous biotic control over resource availability may increase over time (Odum 1969; Leuschner & Rode 1999) leading to linkages between the overstorey and understorey, and increasingly complex hierarchical interactions (Gilliam, Turrill & Adams 1995; Miller, Mladenoff & Clayton 2002; Gilliam 2007). Indeed, spatial patterns of understorey vegetation in primary and old-growth stands are more fine grained and patchy, with lower rates of species accumulation, relative to younger second-growth stands (Scheller & Mladenoff 2002; Burton et al. 2009).

Within forests, fine-scale gradients of soil resources (e.g. moisture, N, P and Ca) and light are generated by multiple factors that interact across scales of space and time. Processes that affect moisture balance and the supply of soil nutrients are tied to exogenous site factors such as soil morphology and topography (Campbell & Norman 1998; Schaetzl & Anderson 2005). Layered upon physiographic factors are endogenous gradients related to the composition and structure of the overstorey, affecting both the radiation regime in the understorey (Canham et al. 1994) as well as the physical and chemical properties of soil at the scale of a single tree (c. 40 m2) to larger patches (>1 ha). In particular, evidence suggests that tree species can exert control over nutrient cycling via litter inputs and uptake (Mladenoff 1987; Ferrari 1999; Lovett et al. 2004; Scharenbroch & Bockheim 2007; Weand et al. 2010). Furthermore, canopy gaps created by wind-throw and tree mortality can alter the local microclimate, resource levels and resource heterogeneity (Moore & Vankat 1986; Mladenoff 1987). The spatial and temporal distribution of such ‘micro-environmental gradients’ can influence patterns of competition among individuals, as well as population and community structure (Hutchings, John & Wijesinghe 2003; Neufeld & Young 2003). Thus, the fine-scale spatial patterning of forest ground-layer plant communities, simply in response to resource gradients, can be quite complex.

Relationships between functional groupings of plant species and environmental gradients can provide evidence for environmental filtering, particularly when the traits suggest an advantage in the associated environment (McGill et al. 2006). Morphology tends to converge among forest herbs with similar leafing phenology suggesting that phenological guild reflects a key ecological strategy (Givnish 1987). For instance, spring ephemerals have relatively short leaf life span and higher rates of photosynthesis and respiration than early summer, late-summer and evergreen species and thus may require greater levels of resources to grow, survive and reproduce (Neufeld & Young 2003; Reich et al. 2003; Diaz et al. 2004). Additionally, many forest herb species have long pre-reproductive periods and short-distance dispersal mechanisms (Whigham 2004). The diversity of ecological strategies suggests the potential for niche differentiation along environmental gradients as well as dispersal limitation and ecological drift (Tilman 1990; Ehrlén & Eriksson 2000; Gilbert & Lechowicz 2004; Ozinga et al. 2005; Gotelli & McGill 2006).

Here we test the hypothesis that fine-scale variation in the composition of ground-layer plant communities is related to species functional groupings and environmental gradients rather than colonization and stochastic drift in a 70- to 90-year-old second-growth northern hardwood forest. Specifically, we expected that the spatial distribution of plant communities within the forest is associated with the effects of interactions between exogenous site characteristics and the live overstorey on the distribution of environmental and resource gradients. We also test the hypothesis that exogenous site variables exert relatively greater control over community composition than do endogenous live overstorey structure variables. Therefore we (i) assess the relative importance of niche partitioning along environmental gradients vs. colonization; (ii) examine how exogenous site characteristics and the overstorey interact to affect the spatial patterning of plant communities in the forest understorey; and (iii) examine the relative importance of exogenous site characteristics vs. endogenous variation related to overstorey composition. We use a structural equation modelling (SEM) framework (e.g. Grace 2006) integrating sub-hypotheses suggested by theory as well as previous investigations (Grace et al. 2010). By examining networks of relationships among ground-layer plant communities, overstorey composition and structure and environmental variables, we quantify the relative importance of niche differentiation and colonization to better understand how the species pool is filtered by local processes in the forest understorey.

Materials and methods

Study area

The study area is located within the Flambeau River State Forest on the loess plain of north-central Wisconsin, USA (Fig. 1). It is a rich, second-growth northern mesic forest (Curtis 1959) and the site of a long-term manipulative experiment that examines the effects of forest structure and deer exclusion on biodiversity and ecosystem function (e.g. Dyer et al. 2010; Stoffel et al. 2010). The field site consists of thirty-five 80 × 80 m (0.64 ha) permanent plots distributed across a 280-ha forest landscape. Each plot contains three circular subplots of 0.05, 0.14 and 0.23 ha. Forty-two permanent 4-m2 quadrats were established in each plot within the small (= 10), medium (= 16) and large (= 16) subplots (Fig. 2). For this study, we focus on pre-treatment data from a subsample of nine plots representative of the variation in overstorey composition and structure encountered at the site. We used a random subsample of 24 quadrats from each of the selected plots (= 4, 7 and 13 from the small, medium and large subplots, respectively; = 215; one plot with 23 quadrats). This subsample was selected for more intensive sampling of environmental variables because it was an economically and logistically feasible sample size.

Figure 1.

 Location of study site in north-central Wisconsin. Inset shows location within the eastern USA.

Figure 2.

 Plot detail with subplot and quadrat layout. Quadrats are distributed along cardinal axes within subplots. Quadrats are 2 × 2 metres (4 m2) and spatially integrate fine-scale measurements of forest structure, microclimate, plant communities and soil properties. Subplot sizes correspond to experimental gap treatments addressed in subsequent studies.

The overstorey includes one major age class of 70- to 90-year-old northern hardwoods dominated by sugar maple (Acer saccharum), basswood (Tilia americana) and white ash (Fraxinus americana), although scattered trees exceeding 100 years of age can be found (Table 1; all nomenclature follows the Flora of North America Editorial Committee 1993+). Soils are silt loams (Glossudalfs) of the Magnor, Ossmer and Freeon series overlaying dense till, all of which are subjected to seasonally perched (Magnor and Freeon) or high (Ossmer) water tables. The Ossmer series is relatively well drained compared to the Magnor and Freeon series, and distributed among lower elevations across the site, although the elevation gradient is subtle (c. 20 m of relief among the subplots in this analysis). To index variability in the local microclimate (e.g. attributable to soil series as well as local topography), subplot elevation, slope and aspect data were gathered from a digital elevation model (30 m resolution). January and July daily high temperatures (1980–1997) average −12 and 19 °C, respectively (Daymet U.S. Data Center, http://daymet.org), and the median length of the growing season is 105 frost-free days (base temperature = 0 °C, 1971–2000, Midwest Regional Climate Center, http://mcc.sws.uiuc.edu).

Table 1.   Overstorey composition (plot mean ± SE of 11 most abundant tree species). Total basal area (BA) is 29 m2 ha−1, density is 445 live stems (≥ 10 cm d.b.h.) ha−1. Foliar chemistry data (mass-based %C, %N and %Ca) are from site (= 3 per species, collected from overstorey tree canopies prior to senescence in September 2008) unless otherwise specified using symbols (superscripts). Foliar P values = 0.2% for all species, except Acer rubrum (for which we do not have data). ND, no data
Percentage by speciesBADensityCNCaLignin
  1. Data are from the †Northeastern Ecosystem Research Cooperative foliar chemistry database (2009) and ‡S. Serbin & P. Townsend, unpubl. data.

Acer saccharum56.2 ± 2.060.3 ± 1.745.21.71.814.8‡
Tilia americana16.0 ± 1.310.4 ± 0.943.52.03.217.0†
Fraxinus americana11.8 ± 1.67.0 ± 1.043.90.92.023.6‡
Carya cordiformis4.3 ± 0.95.0 ± 1.144.22.01.8ND
Fraxinus nigra3.9 ± 0.92.9 ± 0.644.91.21.919.9‡
Quercus rubra2.2 ± 0.60.9 ± 0.247.62.10.922.3‡
Betula alleghaniensis1.6 ± 0.41.1 ± 0.246.12.11.727.9†
Acer rubrum1.4 ± 0.61.6 ± 0.749.9†1.8†ND18.5†
Tsuga canadensis0.9 ± 0.30.5 ± 0.150.51.60.616.0†
Ostrya virginiana0.7 ± 0.23.1 ± 0.844.11.92.621.1†
Populus tremuloides0.6 ± 0.40.6 ± 0.549.42.21.525.6†

Field data collection

Ground-layer vegetation sampling was conducted in the spring (mid-April – late-May) and midsummer (mid-June – late-July) in 2006 to account for all vascular plant species, which vary in leafing phenology. During each survey, the percentage cover of all vascular plant taxa <1.4 m tall was estimated in each 4-m2 quadrat (Fig. 2) using an ordinal cover class: absent, ≤ 0.5%, 0.5–1%, >1–2.5%, >2.5–5%, >5–15%, >15–25%, >25–50%, >50–75%, >75–95%, >95 to < 100%, and 100% (e.g. Gauch 1982). Classes included the range of values that is greater than the lower bound and equal to or less than the upper bound. Seasonal data sets were merged, retaining the largest cover classes when species were observed during multiple sampling periods. Cover classes were converted to mid-points for all analyses. In the midsummer survey, in addition to cover estimates, tree seedlings and saplings (<3 cm d.b.h.) were tallied within each quadrat. All saplings < 10 cm d.b.h. were tallied within subplots. We also measured the diameter of all trees ≥ 10 cm d.b.h. (c. 1.4 m height) within each subplot and calculated basal area as the sum of the stem cross-sectional area of all trees (by species, from d.b.h.) scaled to a consistent per-hectare basis (m2 ha−1) to account for the differences in area among large, medium and small subplots.

Diffuse non-interceptance (%Light) was measured at 30, 100 and 200 cm above each quadrat in 2006 after canopy closure between 31 May and 26 July using the LAI-2000 plant canopy analyzer (LI-COR, Lincoln, NE, USA). The LAI-2000 measures the integrated gap fraction at five concentric angles simultaneously (Welles & Norman 1991). Above-canopy measurements were taken concurrently every 15–30 s on a 31-m tower located centrally within the site. Below-canopy measurements were matched with the closest-in-time above-canopy measurements and %Light was calculated using the post-processing software, FV2000 (LI-COR). All measurements were taken under uniformly overcast or cloudy sky conditions. Trends due to variation over the sampling period and among different sensors were modelled and subsequent analyses relating transmittance to overstorey structure and community data were performed using the residuals.

Percentage volumetric soil water content (SWC) was measured at all quadrats with a time domain reflectometer (TDR; Delta TH20, Dynamax Inc, Houston, TX, USA) to a depth of 6 cm using site-specific calibrations (J. L. Stoffel, unpubl. data). Time domain reflectometry probes were placed in the mineral soil in two locations along the northern boundaries of all quadrats and measurements were averaged. Measurements were taken four times during the 2006 growing season: between 14 and 18 June (June SWC), 11 and 13 July (July SWC), 24 and 28 July (late July SWC), and 14 and 18 Aug. (Aug. SWC). No measurements were taken within 24 h of a heavy rain.

In October (2006), O-horizon samples were collected along four cardinal directions 1 m from quadrats within a 33-cm diameter PVC ring and composited for each location. Corresponding mineral soil samples to a depth of 15 cm were collected using a 2.5-cm diameter push probe. O-horizon samples were dried at 70 °C and soil samples were air dried and passed through a 2-mm sieve. Both O-horizon and soil samples were ground and sent to Brookside Laboratories (New Knoxville, OH, USA) for standard analysis of soil and tissue properties: pH, soil organic matter (OM), sulphur, Mehlich exchangeable Ca, Mg, K and Na, cation exchange capacity (by summation), Mehlich III extractable P, Mn, Zn, B, Cu, Fe and Al for soil, and percentage (%)N, P, K, Ca, Mg, S, B, Fe, Mn, Cu, Zn and Al for the O-horizon. Soil nutrient concentrations were scaled up to a per-hectare basis using regressions of bulk density (BD) on OM (= 119 samples), BD = 0.3215 × OM−0.3434, R2 = 0.99, < 0.001 and calculated assuming a constant coarse fraction. O-horizon N : P ratio was calculated in order to relate ground-layer vegetation to stoichiometric constraints on resource availability (Elser et al. 2000). In the large subplots, additional soil cores were extracted, divided to two depths (shallow = 0–15 cm, deep = 15–50 cm) and submitted for hydrometric soil textural analysis (i.e. %sand, silt and clay).

Dispersal and functional groups

All herbaceous and shrub species (127 species total) were classified (i) according to their life-form as forb, fern, graminoid, shrub or vine; (ii) leafing phenology as spring ephemeral, early summer, late-summer, or evergreen-dimorphic (Givnish 1987); (iii) ability to reproduce vegetatively (clonality) and (iv) seed dispersal mechanism as ballistic, gravity/no obvious mechanism, ingested, myrmecochory, wind or adhesive (Miller, Mladenoff & Clayton 2002). The cover and number of species (richness) within each group was aggregated (by summation) for all quadrats.

Multivariate analysis

To examine relationships among compositional similarity, dispersal and functional groups, and environmental variables we used simple and partial Mantel tests using the ecodist package in R (R version 2.7.2. © 2008. The R Foundation for Statistical Computing; Goslee & Urban 2007). Simple Mantel tests test the null hypothesis of no relationship between two similarity matrices, while partial Mantel tests permit the inclusion of more than two similarity matrices to examine the relationships among correlated variables (e.g. relationship of community composition to geographic distance given the environment). The compositional similarity (Sørenson, or Bray–Curtis, index) of all pairs of quadrats was calculated as twice the sum of the abundances of all species shared between two quadrats divided by the sum of all species abundances occurring in the two quadrats (McCune & Grace 2002). Matrices based on functional or dispersal groupings and environmental variables were calculated in the same manner, except species were replaced, respectively, with the groups and environmental variables. Mantel correlations among the community matrix, the matrix calculated from aggregated functional groups, and the environmental matrix provide evidence of trait-based environmental filtering, while correlations with dispersal group and geographic spatial distance matrices provide evidence for dispersal assembly – particularly in the absence of environmental and functional group correlations. Relationships between community and geographic distances not accounted for by the environmental, functional and dispersal trait matrices were interpreted as evidence of neutral patterns of community assembly (e.g. Tuomisto & Ruokolainen 2006).

To characterize patterns of compositional variation, we used non-metric multi-dimensional scaling ordinations. Non-metric multi-dimensional scaling (NMS) is an ordination technique that iteratively finds the solution, or axes of variation, that best capture the patterns in the dissimilarity matrix (also using the Sørenson measure of dissimilarity). By examining the positions of quadrats along ordination axes (NMS scores), as opposed to cover of individual species, species richness or groupings of species, we can examine the main dimensions of compositional variation in the forest. All species occurring in fewer than 5% of all quadrats were deleted prior to the ordination (80 species deleted) and species cover data were relativized using a general relativization by column (species) totals. We further transformed the main matrix of 47 species using the arcsine square root transformation. Our final solution was based on the rotation that maximized the variation explained along the first axis. Pearson correlations between species abundance and NMS axis scores were examined to interpret how species composition varies along ordination axes. Similarly, we examined the correlations among NMS axis scores and the aggregated cover and richness of all functional and dispersal groups, and environmental variables. Bi-variate plots were assessed visually and data were transformed when necessary. Non-metric multi-dimensional scaling ordinations were executed using PC-ORD (McCune & Mefford 1999).

Structural equation modelling

To examine relationships among overstorey structure, environmental gradients and ground-layer plant community composition (as described by NMS axis scores), we used SEM. Structural equation modelling is based on regression analysis and uses two or more structural equations to examine complex, multivariate hypotheses (Grace 2006). In contrast to multiple regression models, which assume no directional relationships among independent variables, SEMs can better represent the relationships among correlated variables and can lead to different conclusions and improved interpretations (Grace & Bollen 2005). This feature permits one to model direct and indirect interactions among variables, allowing investigators to quantify the relative importance of the multiple pathways by which variables may interact in a system. Structural equation modelling tests the structure of the overall hypothesized model by comparing the implied and observed covariance matrices, as well as the individual ‘paths’ or ‘subhypotheses’ embedded within the model. Although a good fit between data and an SEM by itself may not establish causation, it can provide valuable insight into the processes associated with the pattern of interest when theory and evidence support the hypothesized model.

We postulated an initial conceptual model (meta-model, sensuGrace et al. 2010) based on our previous investigations, theory and a priori knowledge (Fig. 3). In some cases, it could be argued that relationships may be reversed or reciprocal; however, we believe that in the context of this study these premises provide reasonable starting points for an SEM analysis. To relate this theoretical framework to the community axis scores, we replaced conceptual variables (Fig. 3) with measured variables. Measured variables were selected based on the correlations of variables with NMS axis scores. Modification indices were examined, and pathways were added if they improved the overall fit of the model. We selected the best-fitting model that both corresponded with our understanding of the system (Fig. 3) and explained existing patterns of spatial autocorrelation (described below) using Akaike’s information criterion (AIC; Akaike 1974), which favours more parsimonious models. All SEM analyses were performed using amos 18.0 (Amos 18.0.0 ©1983–2009 James L. Arbuckle).

Figure 3.

 Meta-model guiding the structural equation modelling analysis of ground-layer plant communities. Boxes represent categories of variables, while the arrows show the hypothesized relationships among categories and the directions of those relationships. Arrow width indicates hypothesized relative importance of the pathway. Disturbance is represented in a circle and with dotted lines because we do not measure disturbance history directly, but infer its effects from residual spatial patterns indicating dispersal limitation.

Spatial autocorrelation

Spatial correlation of model residuals (SCR) indicates that one of the main assumptions of linear modelling is not satisfied (i.e. independent errors). Additionally, SCR indicates that one or more variables contributing to the spatial patterning of the dependent variable is missing from the model. SCR may be attributed to an unmeasured spatial process such as dispersal limitation or disturbance history (McIntire & Fajardo 2009). To examine the assumption of independent errors, we fit robust semivariogram models (Cressie & Hawkings 1980) to observed NMS axis scores and model residuals (the difference between SEM-predicted NMS scores and observed NMS scores) using weighted least-squares regression. Residual spatial autocorrelation was deemed ‘significant’ when fitted lines occurred outside of the Monte Carlo random simulation envelopes (100 simulations). Spatial analyses were performed in R, using the GeoR package (R version 2.7.2. © 2008. The R Foundation for Statistical Computing).

Results

Associations among community similarity, environment and traits

Mantel tests showed that the similarity matrix calculated from species abundances (community composition) is related to matrices calculated from aggregated dispersal and functional groupings, environmental variables and geographic distance (Table 2). While the similarity matrix based on dispersal groupings is significantly related to the matrix based on functional groups (r = 0.43, < 0.001), both simple and partial Mantel tests suggest that dispersal groups are more strongly related to community composition than functional groups (Table 2). Partial Mantel tests relating the community matrix to the environmental matrix and geographic distance matrix also show stronger relationships of community composition to geographic distance (i.e. space) than environmental variables.

Table 2.   Simple and partial Mantel tests examining correlations among similarity in species composition, dispersal and functional traits, environmental variables and spatial coordinates
Community–Trait RelationshipsMantel rP
  1. The community matrix was calculated from the relative abundances of all species occurring in > 5% of the quadrats (47 species); the functional group matrix was based on the aggregated cover of life-forms (ferns, forbs, graminoids, shrubs and vines) and phenolological guilds (spring ephemeral, early summer, late-summer and evergreen-dimorphic); the dispersal matrix was calculated based on the aggregated cover of species with adhesive, ballistic, clonality, ingested, myrmecochory or wind seed and vegetative dispersal. The environmental matrix included all exogenous site variables; endogenous overstorey tree species and saplings [basal area (BA) and density, respectively – saplings not separated by species]; and environmental variables = soil and O-horizon properties and light transmittance.

 Community Dispersal0.42<0.001
 Community Functional0.35<0.001
 Dispersal Functional0.43<0.001
 Community Dispersal|Functional0.32<0.001
 Community Functional|Dispersal0.21<0.001
Community–Environment Relationships
 Community Environment0.25<0.001
 Community Space0.32<0.001
 Environment Space0.37<0.001
 Community Environment|Space0.15<0.001
 Community Space|Environment0.25<0.001

Ordination

The three-dimensional NMS ordination of 215 quadrats (Fig. 4) has a final stress of 23.86 and explains a total of 54.7% of the variation in the ground-layer plant community composition and structure (axis 1 = 23.7%, axis 2 = 20.3%, axis 3 = 10.7%). NMS ordinations with stress levels > 20 are typically considered more difficult to interpret; however, stress is strongly related to sample size (McCune & Grace 2002) and our sample of 215 quadrats is unusually high for community studies. Monte Carlo randomization tests suggest that observed patterns were unlikely due to chance (α = 0.05). Furthermore, repeated ordinations as well as ordinations with smaller samples, greater explained variability and lower stress levels yield consistent patterns.

Figure 4.

 Non-metric multi-dimensional scaling (NMS) ordination of quadrats (= 215) described by 47 species. Vectors from centroids show joint axis correlations with dispersal (dashed lines, italic font) and functional groups (solid lines, phenological guilds shown in bold font; r ≥ 0.40). Axis labels show the percentage of variation in the distance matrix captured by the ordination axis and main species correlations (Table 3). Species listed along vertical axes include Viola pubescens, Arisaema triphyllum, Athyrium filix-femina and Ribes cynosbati. ES, early summer; FRB, forb; CLN, clonal; AD, adhesive; BAL, ballistic; GRV, gravity; SE, spring ephemerals; and EV, evergreen-dimorphic; sp, species richness; cov, percent cover.

The first axis contrasts communities characterized by greater coverage and richness of spring ephemerals from communities composed primarily of evergreen-dimorphic species (Table 3). The functional and dispersal groups positively associated with axis 1 include cover and richness of spring ephemeral species (r = 0.46 and 0.64, respectively), cover and richness of ballistic species (r = 0.50 and 0.52, respectively), cover of gravity-dispersed species (r = 0.49) and cover of forb species (r = 0.40), while cover and richness of evergreen-dimorphic species (r = −0.42 and −0.43, respectively) are negatively related to quadrat axis 1 scores (Fig. 4, Table 3).

Table 3.   Correlations of species cover with ordination axis scores (by phenological group). Only species with Pearson axis correlations (r) ≥ 0.40 are given
 Axis 1Axis 2Axis 3
Spring ephemerals
 Allium tricoccum0.560.29−0.12
 Enemion biternatum0.460.09−0.12
 Dicentra spp.0.000.480.18
Early-summer species
 Viola pubescens−0.130.630.06
 Trientalis borealis−0.48−0.110.07
 Ribes cynosbati−0.020.02−0.50
 Polygonatum pubescens−0.480.380.08
 Phlox divaricata−0.070.450.10
 Osmorhiza spp.−0.170.530.05
 Athyrium filix-femina0.01−0.120.50
Late-summer species
 Laportea canadensis0.080.120.47
Evergreen-dimorphic species
 Dryopteris intermedia−0.40−0.230.26
 Anenome acutiloba−0.46−0.150.28

Axis 2 contrasts diverse communities of early summer forbs from sparsely vegetated communities composed of less common graminoid and evergreen-dimorphic species (Fig. 4, Table 3). Total plant species richness (r = 0.47) and richness of herbaceous plant species (r = 0.47) are positively correlated with axis 2. Forb richness and cover (r = 0.60 and 0.51), early summer species cover and richness (r = 0.40 and 0.59, respectively), richness of clonal species (r = 0.49) and adhesive species richness and cover (r = 0.51, for both) are positively correlated with axis 2, while no major groups are significantly negatively related to axis 2 (r cut off = 0.40).

The third axis distinguishes communities of fern and late-summer species from communities composed of shrub species (Fig. 4, Table 3). Fern cover (r = 0.52) and fern species richness (r = 0.45) are positively related to axis 3 while shrub species richness (r = −0.47) and shrub cover (r = −0.47) are negatively related to axis 3.

Structural equation modelling

The final structural equation models showed correspondence between the implied and observed covariance matrices (χ2 = 1.04, d.f. = 3, = 0.793 for axis 1; χ2 = 8.73, d.f. = 7, = 0.273 for axis 2). No model was fit to axis 3 because Pearson correlations indicate only very weak associations with environmental variables. In SEM analysis, P-values ≥ 0.05 indicate a fit between the data and the model; therefore our P-values in addition to other fit indicators, such as AIC, suggest a good model fit. In order to simplify the structural equation models, we used statistical composites (Verheyen et al. 2003; Grace 2006; Harrison et al. 2006) for overstorey composition, soil texture, O-horizon properties and soil properties (chemical). Observed indicator variables for each composite were selected based on correlations with NMS axis scores and backwards elimination multiple regression analysis. Composites were constructed for each axis using generalized linear models (SAS® GLM Procedure) and predicted scores were used as soil texture, O-horizon, and soil properties variables (Fig. 5).

Figure 5.

 Composite variable models relating observed soil texture, chemical properties and O-horizon to NMS axis scores (Fig. 4). Solid lines show positive relationships, dashed lines show negative relationships; line weights correspond to the variable importance. Standardized coefficients and significance levels (*< 0.05, **< 0.01, ***< 0.001) are shown along the paths. %Siltdeep, %Silt in deep horizons (15–50 cm); %Claydeep, %Clay in deep soil horizons; %Claysurface, %Clay in shallow soil horizons (0–15 cm); N : P, ratio of nitrogen to phosphorous in O-horizon; %Ca, percentage calcium in O-horizon; pH, soil pH; CEC, soil cation exchange capacity; Ln, indicates a natural log transformation; P, soil phosphorus; Ca, calcium; O.M., soil organic matter; see Table 1 for full tree species nomenclature.

A hierarchy of controls driven by soil texture and associated effects on soil moisture, chemical properties of soil, and overstorey composition and structure, explains 62% of the total variation in quadrat axis 1 scores (Table 4, Fig. 6). While axis 1 scores are directly related to elevation, July SWC, O-horizon and soil properties, these direct effects are mediated by soil texture and the composition of the overstorey. Given these direct and indirect effects, elevation has the strongest total effects on axis 1 (standardized total effect = 0.66), followed by soil properties (standardized total effect = 0.26) and July SWC (standardized total effect = 0.24). The total contributions of O-horizon properties (standardized total effect = 0.14) and the overstorey (standardized total effect = 0.16) were relatively weak.

Table 4.   Structural equation model (SEM) estimates. Arrows show direction of path from variables to the left to variables to the right of the arrow. Estimates are unstandardized coefficients, z is the test statistic, and pr ≥ |z| is the two-tailed probability under the null hypothesis of no relationship between variables. Standardized coefficients (Std. Coeff.) are standardized by standard deviations. For axis interpretations see Fig. 4 and Table 3
AXIS 1
 Estimatezpr ≥ |z|Std. Coeff.
Direct effects
 O-horizon → Axis 10.182.010.0440.10
 Soil properties → Axis 10.406.39<0.0010.26
 Elevation → Axis 10.056.45<0.0010.38
 July SWC → Axis 10.043.99<0.0010.20
Indirect effects
 Elevation → July SWC0.246.35<0.0010.42
 Elevation → Overstorey0.0715.31<0.0010.77
 Elevation → O-Horizon0.035.06<0.0010.44
 Soil Texture → July SWC2.342.200.0280.15
 Soil Texture → Overstorey−0.48−4.25<0.001−0.18
 Soil Texture → Soil Properties0.433.61<0.0010.20
 Soil Texture → O-Horizon−0.38−3.280.001−0.19
 July SWC → Overstorey0.033.270.0010.16
 July SWC → O-Horizon0.022.690.0070.16
 Overstorey → O-Horizon0.193.070.0020.25
 Overstorey → Soil Properties0.406.83<0.0010.47
 O-Horizon → Soil Properties0.162.060.0390.14
Unexplained correlations
 Soil Texture ↔ Elevation0.405.85<0.0010.44
 Axis 1 ↔ Overstorey0.055.61<0.0010.44
AXIS 2
 Estimatezpr ≥ |z|Std. Coeff.
  1. July SWC = soil water content measure between 24 and 28 July, June SWC measured between 14 and 18 June. For construction of Overstorey, Soil Texture, Soil Properties and O-Horizon Properties composite variables see Fig. 5.

Direct effects
 O-Horizon → Axis 20.726.24<0.0010.38
 June SWC → Axis 2−0.02−2.370.018−0.13
 sqrt(Gravity) → Axis 20.063.83<0.0010.21
 sqrt(Ballistic) → Axis 20.072.930.0030.17
 Soil Texture → Axis 20.442.110.0350.12
Indirect effects
 Soil Texture → June SWC−4.43−2.740.006−0.18
 Soil Texture → Overstorey0.857.13<0.0010.44
 Soil Texture → O-Horizon−0.44−3.50<0.001−0.23
 June SWC → O-Horizon−0.02−4.25<0.001−0.25
 Overstorey → O-Horizon0.528.24<0.0010.53
 O-Horizon → Soil Properties0.173.170.0020.20
Unexplained correlations
 Axis 2 ↔ Overstorey0.033.53<0.0010.26
 sqrt(Gravity) ↔ sqrt(Ballistic)0.884.19<0.0010.29
 sqrt(Gravity) ↔ Overstorey0.092.450.0140.16
 sqrt(Ballistic) ↔ Overstorey0.124.36<0.0010.30
 O-Horizon ↔ sqrt(Gravity)−0.10−2.870.004−0.19
 Soil Texture ↔ sqrt(Gravity)0.093.83<0.0010.27
 Soil Texture ↔ sqrt(Ballistic)0.084.40<0.0010.32
 sqrt(Ballistic) ↔ Soil Properties0.052.480.0130.15
Figure 6.

 Results of structural equation modelling analysis (χ2 = 1.04, d.f. = 3, = 0.793, AIC = 67 for axis 1 and χ2 = 8.73, d.f. = 7, = 0.273, AIC = 63 for axis 2). Straight, single-headed arrows show directed effects of one variable on another; curved, double-headed arrows indicate unanalysed correlations between variables. Dashed lines are negative relationships and solid lines show positive relationships; weights correspond with the strength of the standardized coefficients (Table 4). R2 values show the proportion of variance explained for each endogenous variable by all variables with direct arrows to it. Composite variables (Figs 4 and 5) are indicated within hexagons; observed variables are in rectangles. Unanalysed correlations among the aggregated cover of species with ballistic and gravity seed dispersal, the overstorey and environmental variables in the model for axis 2 are not shown graphically. SWC, soil water content; Elev., elevation.

Interactions among soil texture, overstorey composition, soil moisture, O-horizon properties and dispersal traits explain 41% of the variation in axis 2 scores (Table 4, Fig. 6). Axis 2 scores are directly related to soil texture, soil moisture (June), O-horizon properties and aggregated dispersal traits. Specifically, the aggregated cover of species with gravity and ballistic seed dispersal resolved patterns of residual spatial autocorrelation that could not be explained by environmental variables. Additionally, June SWC and soil texture have indirect effects on axis 2 through their effects on other variables directly related to axis 2. O-horizon properties (standardized total effect = 0.38), followed by gravity seed dispersal (standardized total effect = 0.22), June SWC (standardized total effect = −0.22) and overstorey composition (standardized total effect = 0.20) have the strongest influences on axis 2.

Discussion

Our first objective was to assess the relative importance of environmental filtering vs. colonization processes for fine-scale patterns of forest ground-layer plant communities in this second-growth northern mesic forest. Results suggest that both processes play important roles in plant community assembly. The spatial patterning of ground-layer plant communities is often attributed to corresponding changes in environmental conditions and dispersal limitation (Nekola & White 1999; Miller, Mladenoff & Clayton 2002; Gilbert & Lechowicz 2004). Although stochastic processes resulting in ecological drift can produce strong levels of dispersal limitation (Hubbell 2001), the relationship between dispersal strategies and community patterns that we observed suggests that these spatial patterns are not generated solely by stochastic processes of ecological drift. In particular, dispersal limitation appears relatively more important for species with short-distance dispersal mechanisms (i.e. gravity or ballistics) than species with long-range seed dispersal (i.e. wind, adhesion or ingestion). Ozinga et al. (2005) observed similar colonization constraints along environmental gradients spanning the entire biogeographic region of the Netherlands. This pattern, in conjunction with the observed correlations among community-aggregated dispersal and functional groupings, suggests that competition–colonization trade-offs (Tilman 1990), or tolerance–fecundity trade-offs associated with seed size (Ehrlén & Eriksson 2000; Muller-Landau 2010), may play an important role in the spatio-temporal patterning of ground-layer plant communities. Furthermore, our results suggest that nearly a century after the original logging disturbance, ground-layer plant communities are not at competitive equilibrium with the environment.

Patterns of compositional variation can be difficult to detect in the forest understorey, particularly at the fine grain and large extent examined here (i.e. = 215, 4-m2 quadrats distributed across a 280-ha research site). Using ordinations of quadrats in species space, we identified three dimensions of compositional variation that interact to produce the observed mosaic of plant communities (Fig. 4, Table 3). The transition from evergreen-dimorphic to deciduous and spring ephemeral species along axis one has been seen elsewhere (Scheller & Mladenoff 2002; Graves, Peet & White 2006; Burton et al. 2009), and suggests that phenological guilds not only permit species to partition temporal variability in the environment (e.g. Givnish 1987), but also spatial gradients in soil nutrients (mainly phosphorus, in this case). The pattern of environmental sorting may reflect trade-offs between survival in stressful, resource-poor environments and growth in resource-rich environments (e.g. Reich et al. 2003; Diaz et al. 2004). The stronger relationship of spring ephemerals rather than evergreen-dimorphic species with community axis scores may reflect the relatively high whole-plant compensation points of spring ephemerals (Givnish 1987). By contrast, evergreen-dimorphic species appear in a broader range of environmental conditions.

The orthogonal patterns of increasing cover of early summer species contrasting with less common graminoid species and forbs such as Arisaema triphyllum, albeit modest, is similar to patterns shown with gradients in the severity of exotic earthworm invasion in forests of the northern Great Lakes region (e.g. Frelich et al. 2006). After accounting for environmental influences, patterns of residual spatial autocorrelation are related to the abundance of species with short-distance ballistic and gravity seed dispersal mechanisms (Fig. 6). This ‘colonization deficit’ (sensuOzinga et al. 2009) cannot be attributed to changes in the availability of dispersal vectors because these two mechanisms do not rely upon external agents or fragmentation in this contiguous forest (but see Damschen et al. 2008). It is more likely associated with lower rates of colonization for these species and the fine-scale connectivity of suitable microsites (Ehrlén & Eriksson 2000) in this contiguous forest. Additionally, it is possible that environmental variables such as soil moisture and O-horizon %Ca increase earthworm activity (Hobbie et al. 2006) to result in local extirpation of such species (Hale, Frelich & Reich 2005), producing the observed spatial patterns. Interactions with herbivores, such as white-tailed deer, may further exacerbate this problem (Frelich et al. 2006). Future studies will examine this hypothesis directly.

Environmental filtering

We hypothesized that interactions among exogenous site characteristics and overstorey structure, through their effects on the distribution of resources within the forest, affect the spatial patterning of plant communities in the forest understorey. Thus, embedded within the overall hypothesis that such interactions exist were several sub-hypotheses related to the nature of the interactions among environmental variables (Fig. 3). Because some of our sub-hypotheses were not supported by the data, our results suggest a more complex version of the original conceptual model. The main differences between the original and revised meta-models are (i) a pathway from soil moisture to O-horizon properties; (ii) a pathway from O-horizon properties to ground-layer plant communities; and (iii) a lack of important pathways to and from light transmittance. The paths from soil moisture to O-horizon properties likely reflect the positive relationship between soil moisture and nitrogen and phosphorus mineralization in well-drained soils (Fig. 6; Mladenoff 1987) and denitrification in saturated soils (Fig. 6; Christensen, Simkins & Tiedje 1990). Similarly, the paths from O-horizon properties to plant communities likely indicate the important effects of litter quality on the rates at which nutrients become available to plants for uptake (Ferrari 1999; Lovett et al. 2004). However, because this study covered a single growing season, results from modelling may be different in another year, although we expect a similar structure.

Because of the important role of canopy gaps in temperate deciduous forests (e.g. Moore & Vankat 1986) and because light is generally thought to be the most important limiting resource in forest understoreys (Neufeld & Young 2003), we expected light transmittance to be related to the spatial patterning of ground-layer plant communities (Graves, Peet & White 2006). We also expected light transmittance to be related to the composition and structure of the overstorey (Canham et al. 1994), although we observed neither pattern. The importance of light gradients may be restricted to stands with a greater range of values, including lower minimum and higher maximum transmittances, such as old growth stands with canopy gaps created by the fall of old, large trees (Dahir & Lorimer 1996). For instance, Scheller & Mladenoff (2002) observed that the variability of light transmittance within a stand was associated with differences in the spatial patterning of understorey plant communities between old-growth and mature second-growth stands. Old-growth stands exhibit greater levels of within-stand heterogeneity in light transmittance than second-growth stands (CV = 65% vs. 32%) and a corresponding finer-grained mosaic of ground-layer plant communities. The variability in light transmittance here is more similar to that in the even-aged stands of Scheller & Mladenoff (2002) than the old growth (CV = 42% in the present study). Thus, linkages between overstorey and ground-layer vegetation due to canopy gaps may also be restricted to older stands (Gilliam, Turrill & Adams 1995). Indeed, northern hardwood stands can exist in the mature, self-thinning phase for 100–150 years before transitioning to uneven-aged status (Frelich 2002). The timing of this transition likely depends upon the rate of autogenic succession and the occurrence of disturbance, which can accelerate or hinder the development of stand structure.

It is also possible that ground-layer plant communities in younger stands respond to variation in light transmittance induced by variation in leafing phenology among tree species early in the growing season (e.g. Lopez et al. 2008). For instance, although hemlock trees did not appear to influence light transmittance at the peak of the growing season (maximum leaf area index), the reduction in light transmittance in the neighbourhood of hemlock trees in the spring may be another pathway by which hemlock discourages spring ephemeral species with high whole-plant compensation points. Alternatively, current instrumentation available for measuring light transmittance may not be sensitive enough to detect subtle variation in light levels in deeply shaded understoreys (less than c. 6%; Machado & Reich 1999; Tobin & Reich 2009). Controlled and manipulative studies examining limitations of instrumentation and the role of light transmittance integrated over the course of the growing season in structuring understorey plant communities are required to better understand the importance of light in a dimly lit forest understorey.

Exogenous variables

The important direct and indirect effects of site variables and soil moisture in both structural equation models provides confirmatory evidence for the hypothesis that exogenous variables exert control over resources and ground-layer plant communities in this 70- to 90-year-old second-growth forest. However, the relative importance of exogenous vs. endogenous processes varied between the two axes modelled, suggesting that the distribution of resources affecting ground-layer plant communities is not exclusively driven by exogenous site variables. The effects of site variables, and elevation in particular, on soil and O-horizon properties were more important than overstorey composition for axis 1, while the effects of overstorey composition on the O-horizon was cumulatively more important for axis 2. Our results are consistent with those of Smith et al. (2008), who showed that in the absence of disturbance, forest understorey vegetation was related to soil properties and drought. Leuschner & Lendzion (2009) also reported that the soil microclimate and chemical properties were of primary importance to the distribution of forest herbs, while light transmittance was of secondary importance. It remains unclear, for the present study, whether the effects of moisture levels on axis 1 were related to drought stress and/or the effects of moisture on nutrient acquisition (i.e. low levels of soil moisture decrease nutrient mineralization rates, mass flow and diffusion, increasing the importance of root interception for nutrient acquisition; Lambers, Chapin & Pons 1998).

Endogenous variables

O-horizon and soil properties driving patterns of ground-layer plant communities were also related to the composition and structure of the live overstorey. While effects of tree species on soil resource gradients through their differences in litter chemistry have been observed elsewhere (Ferrari 1999; Scharenbroch & Bockheim 2007; Weand et al. 2010), few studies have simultaneously shown how these effects can affect forest understorey plant communities (e.g. Beatty 1984). Based on these results, we propose that the effects of overstorey composition and structure on soil resources is an additional mechanism for observed linkages between overstorey and understorey vegetation later in succession (e.g. Gilliam, Turrill & Adams 1995; Gilliam 2007). Such linkages are likely diminished in younger stands and particularly in second-growth stands due to the decreased importance of foundational species, such as hemlock and other conifer species, in second-growth northern hardwood stands (e.g. Schulte et al. 2007). However, the soil legacy of a greater abundance of these species historically may be causing some of the observed patterns that cannot be linked to the overstorey. For instance, the effects of Roman agriculture on soils and plant communities in forests that were last farmed nearly two millennia ago remain detectable (Dupouey et al. 2002).

O-horizon and soil properties related to the cycling of nitrogen and phosphorus (i.e. O-horizon N:P, soil P, soil and O-horizon Ca) were the most important environmental variables driving the observed compositional patterns. Local soil nutrient status can affect competitive hierarchies, herbivory, soil microbial communities (i.e. mycorrhizae, pathogens) and promote exotic species invasion (Gilliam 2006). Interestingly, nitrogen and phosphorus were associated with different, orthogonal patterns of compositional variation. O-horizon and soil properties associated with the nitrogen cycle interacted with the soil moisture regime to affect the distribution of early summer species. Thus, reduction of heterogeneity in O-horizon and soil properties through the simplification of overstorey composition and structure coupled with nitrogen deposition could result in corresponding homogeneity in ground-layer plant communities (Gilliam 2006; Bobbink et al. 2010).

A transition from communities of evergreen-dimorphic species and spring ephemerals was associated with a gradient in soil phosphorus. Soil phosphorus becomes progressively depleted with substrate age, thus phosphorous limitation is generally thought to be less important in northern temperate zones than in older, weathered soils of the tropics (Walker & Syers 1976; Vitousek & Farrington 1997; Lambers et al. 2008). However, our results, along with the results of Naples & Fisk (2009), suggest that vegetation, or perhaps the distributions of some species, in northern temperate forests may be limited by phosphorous (Weand et al. 2010). Furthermore, this 70- to 90-year-old forest includes scattered trees > 100 years of age, and this oldest age class suggests the site was not completely clear-cut and heavily burned, as often occurred in this region following logging of original stands. Therefore, our study also suggests that P-limitation may not be restricted to earlier stages (c. 30 years old) of stand development (Naples & Fisk 2009). Phosphorus was the most variable soil property that we observed (range = 9.5–220.4 kg ha−1, CV = 79%). The relationship between soil P and spring ephemeral cover may explain the observations of Rogers (1982), who observed that across the entire Lake States region, vernal herb communities were sparser and less diverse on sites with coarse-textured, nutrient-poor soil with slow rates of decomposition and evergreen trees.

Unexplained correlations

While many of the relationships among the overstorey, soil and O-horizon properties and ground-layer plant communities could be specified in the model, the interactions are complex and several additional ‘unanalysed’ correlations had to be controlled for in the model. Unanalysed correlations may be spurious, but they may also indicate a common unobserved driver. For example, a significant unanalysed correlation between overstorey composition and understorey plant community NMS axis 1 scores was observed (Fig. 6). This relationship may indicate an additional unobserved pathway by which these species affect ground-layer plant communities (e.g. overstorey phenology, disturbance history or soil microbial communities).

Conclusion

Our results suggest that the spatial patterning of ground-layer plant communities in this 70- to 90-year-old, second-growth northern hardwood forest is driven jointly by colonization processes and environmental filtering. Although it has been nearly a century since the last major disturbance to this forest (near-complete logging), it appears that ground-layer plant communities have not achieved competitive equilibrium. While it is also possible that competitive equilibrium is a rare state under the natural disturbance regime, and that ground-layer plant meta-communities exist in a dynamic equilibrium resulting from fluctuating environmental conditions (e.g. Connell 1978; Hanski 1998; Leibold et al. 2004), the observed patterns of colonization are not related to gap dynamics. The speed of re-colonization likely depends on the severity of disturbance to the overstorey, understorey (amount of vegetation removed) and forest floor (Roberts 2007). We do not possess specific records of the logging operations at this site; however, disturbance to the overstorey was often severe during the logging era of the early 1900s, leaving residual stands of saplings and small trees (poles), and was often followed by slash fires. The observed even-aged structure, with the exception of a few older trees > 100 years suggests that this was the case here. Areas surrounding residual poles may have served as refugia for slowly colonizing plants – disturbance was likely less severe and competition with shrubs and saplings may be suppressed in such areas. In addition to recovering from the original logging event more slowly, species with short-distance dispersal may be disproportionately affected by exotic earthworm invasions and herbivory by overpopulated white-tailed deer (Odocoileus virginianus) as a result of their low migration rates.

Fine-scale variation in plant communities was also related to environmental filters resulting from complex interactions among site variables, the composition of the overstorey, and O-horizon and soil properties. Much of the variation in soil and O-horizon properties driving ground-layer plant communities is related to exogenous site variables such as soil texture and soil moisture. Relative to exogenous variables, endogenous processes played an important but lesser role in regulating the distribution of resources, and thus ground-layer plant communities. Relationships among tree species, O-horizon and soil properties may have been more important historically when co-dominant species (i.e. hemlock and yellow birch) were locally more abundant. The relative importance of endogenous processes in second-growth stands is thus governed by the composition of the residual stand, the distribution of such biological and ecological legacies and the rate of structural development. Because internal regulation by endogenous processes may be diminished in such second-growth stands due to the removal of coarse woody debris and the selective removal of tree species, they may be more sensitive to climate change and other external stressors than primary forests that were never logged and old-growth forests with higher levels of heterogeneity. Long-term manipulative field experiments coupled with physiological studies of plant resource economics, along with strictly confirmatory SEM analyses performed on independent data sets are required to confirm the causal mechanisms discussed.

Acknowledgements

Funding was provided by grants from USDA McIntire-Stennis to D.J.M. and J.I.B.; the National Research Initiative of the USDA Cooperative State Research, Education and Extension Service to D.J.M., C.G. Lorimer and S.T. Gower; Wisconsin Department of Natural Resources (WDNR) Division of Forestry & Bureau of Integrated Science Services to D.J.M. and J.A.F.; and the Garden Club of America Fellowship in Ecological Restoration to J.I.B. We gratefully acknowledge the contributions of J.H. Dyer, T.D. Hayes, E.F. Latty, J.L. Stoffel and numerous undergraduate assistants; as well as D. Hvizdak and A. Voightlander (U.S. NRCS) and E. Padley (WDNR) for classifying the soil profiles. For feedback on earlier versions, we thank E.I. Damschen, T.J. Givnish, S.T. Gower, E.L. Kruger, C.G. Lorimer, A.E. Sabo, P.A. Townsend, the Handling Editor and two anonymous referees.

Ancillary