Study site and field sampling
The Three Bend Scenic and Wildlife Management Refuge Area is part of Oak Ridge National Environmental Research Park near Oak Ridge, TN (35-58′N, 84-17′W). The Three Bend Area consists of a mix of hardwood forests and old fields. Old-field communities were agricultural fields until abandonment ca. 1943. Soils at the sites are characterized as Typic Hapludult with a silty clay loam texture. Mean monthly temperatures range from approximately 3 °C in the winter to 31 °C in the summer and mean rainfall is 1322 mm.
Sampling biotic and abiotic variables
In the summer of 2006, we sampled seventeen old fields ranging from ca. 2000 to 50 000 m2. We chose these fields based on the presence of well-defined boundaries such as forests or roads. We randomly placed 50-m transects in each field (2–6 transects depending on field area) (Table S1). Along each transect, we placed five 1-m2 plots 10 m apart starting 10 m from the origin of each transect. Hereafter, we refer to the data from each 50-m transect as local scale, and landscape scale refers to all transects in a single field.
In each 1-m2 plot, we identified all plant species (Table S2), tallied exotic and native species richness, and percent foliar cover of all vascular plant species during the peak of the growing season. We also estimated above-ground biomass by randomly placing 0·5 m × 1 m subplots within each 1-m2 plot and clipping all individuals rooted inside to approximately 1 cm from the soil surface. We sorted the biomass into total above-ground biomass (i.e. live plant material) and litter mass (i.e. dead plant material) and then oven-dried the biomass samples for 48 h at 65 °C and weighed them. We also estimated light availability, soil moisture and soil properties (soil pH, soil texture, soil N) in each of the 1-m2 plots (Table S3). To estimate light availability, we recorded photosynthetic photon flux density with a line-integrating ceptometer (Decagon Accupar; Decagon Devices, Pullman, WA, USA) positioned horizontally at approximately 2 cm above the ground in each 1-m2 plot. To measure soil moisture, we used a hand-held time domain reflectometer with 12-cm probes (Hydrosense; Decagon Devices) in one random location per 1-m2 plot to estimate percent volumetric water content. We also collected a 10-cm soil core from the centre of each 1-m2 plot to quantify soil texture (percent sand and clay), bulk density, gravimetric water content, pH and potential net nitrogen ( and ) mineralization. To estimate potential net nitrogen ( and ) mineralization (i.e. potential soil N available for plant uptake), we incubated soil subsamples from each 1-m2 plot for 1 month and compared nitrogen availability of the incubated subsamples with that of subsamples extracted prior to incubation. Soil nitrate () and ammonium () in samples were extracted by 2 m KCL and analysed with an autoanalyzer (Lachat Quikchem 8000; Hach Corporation, Loveland, OH, USA).
Estimating landscape variables
We mapped the boundary of each old field with GPS units and used ARCGIS 9.1 (ESRI, Redlands, CA, USA) to calculate field area and perimeter, density of roads within a 250-m buffer from the edge of each field, and type and area of land cover (other fields, forest, water) within a 250-m buffer from the edge of each old field (Table S3). We calculated road density using the Anderson County roads layer from the Tennessee Spatial Data Server (http://www.tngis.org) and retrieved land cover data from the most recent National Land Cover Database (NLCD 2001; http://www.mrlc.gov). We calculated field edge using Patton’s Shape Index (Patton 1975) and used field measurements of slope and aspect to calculate heat load (McCune & Keon 2002), an integrative measure of the field exposure to incident sunlight for each field (Table S3). We created four categories of mowing frequency based on Tennessee Wildlife Resource Agency records (J. Evans, pers. comm.) as our disturbance measure. We assigned each field to a mowing intensity ranging from 1 to 4, with one representing monthly mowing and four representing biennial mowing.
To examine the NERR at local and landscape scales, we performed linear regressions using cumulative native plant richness to predict cumulative exotic plant richness across 50-m transects (n = 50) and old fields (n = 17) (Table S3). In addition, we assessed variation in the NERR among old fields using a similar sampling approach to that of Davies et al. (2007), regressing exotic plant richness against native plant richness for the 1-m2 plots within each of the 17 old fields.
We also examined the support for the favourable environment or spatial heterogeneity hypotheses as influences on NERR within old-field communities. To test the favourable environment and spatial heterogeneity hypotheses, we used a stepwise linear regression with the slope of NERR (generated across 1-m2 plots within each old-field community) as a continuous response variable and the mean and variation (estimated as the coefficient of variation) in biotic (total above-ground biomass and foliar cover) and abiotic (soil VWC, soil pH, soil N, soil bulk density, litter mass) variables at the landscape scale as potential predictor variables. Prior to regression analyses, we created a correlation matrix among mean and variance in biotic and abiotic factors to assess potential covariation among factors. We tested for significant correlations between all predictor variables using Pearson’s correlation coefficients. Predictor variables with significant correlation coefficients (−0·75 > r > 0·75) were not used in the same model (Kumar, Stohlgren & Chong 2006). We generated NERR slopes, the correlation matrix and the multiple linear regressions with JMP 6.0 (SAS Institute Inc., Cary, NC, USA).
To elucidate which factors might influence exotic plant richness and assess whether those factors varied across spatial scales, we conducted variable selection procedures using all possible regression methods at both local and landscape scales. At both local and landscape scales, we included the measured biotic (native plant richness, total above-ground biomass, exotic cover) and abiotic variables (light availability, soil moisture, soil N, soil bulk density, soil texture, litter mass, heat load) as well as landscape variables (field density, forest density, road density, field edge, mowing regime; Table S3) in our model selection procedure. To estimate biotic variables at the transect and old-field scale (e.g. total species richness, above-ground biomass, etc.), we summed values from all of the 1-m2 plots in each transect and old field. To estimate abiotic variables (e.g. light availability, soil N), we averaged the values from each 1-m2 plot for each transect and old field.
We used the Akaike Information Criterion adjusted for small sample size (AICc; Burnham & Anderson 2002) to evaluate multiple regression models predicting exotic species richness at the local and landscape scales. We tested for collinearity among biotic (native plant richness, total above-ground biomass, exotic cover), abiotic (light availability, soil moisture, soil N, soil bulk density, soil texture, litter mass, heat load) and landscape (field density, forest density, road density, field edge, mowing regime) variables at the landscape scale as potential predictor variables. We tested for collinearity among biotic, abiotic and landscape predictors using the same procedure employed during NERR analyses. All regression analyses were performed using SAS 9.1.3 (SAS Institute, Inc.).
We used Moran’s Istd correlograms to test for spatial autocorrelation in the residuals of the best models (based on biotic, abiotic and landscape predictors) for exotic and native plant species richness at the local and landscape scales. If we found significant autocorrelation in the environmental model residuals, we constructed spatial models and environment + spatial models to account for this autocorrelation (Borcard & Legendre 2002; Borcard et al. 2004). Finally, we found no evidence of spatial autocorrelation in exotic richness at either local or landscape scales (Fig. S1), so we do not discuss it further.