Site conditions are more important than abundance for explaining plant invasion impacts on soil nitrogen cycling

Invasive plant species can alter critical ecosystem processes including nitrogen transformations, but it is often difficult to anticipate where in an invaded landscape, these effects will occur. Our predictive ability lags because we lack a framework for understanding the multiple pathways through which environmental conditions mediate invader impacts. Here, we present a framework using structural equation modeling to evaluate the impact of an invasive grass, Microstegium vimineum (M.v.), on nitrogen cycling based on a series of invaded sites that varied in invader biomass and non-M.v. understory biomass, tree basal area, light availability, and soil conditions. Unlike previous studies, we did not find an overall pattern of elevated nitrate concentrations or higher nitrification rates in M.v.-invaded areas. We found that reference plot conditions mediated differences in mineralization between paired invaded and reference plots at each site through indirect (via M.v. biomass), direct, and interactive pathways; however, the strongest pathways were independent of M.v. biomass. For example, sites with low reference soil nitrate and high non-M.v. understory biomass tended to have faster mineralization at 5–15 cm in invaded plots. These findings suggest that more attention to reference conditions is needed to understand the impact of invasive species on soil nitrogen cycling and other ecosystem processes and that the greatest impacts will not necessarily be where the invader is most abundant.

Table S1.List of reference environmental variables included in the full model to explain Δ nitrate, Δ ammonification, Δ nitrification, and Δ mineralization (Δ = Inv.-Ref.).An "x" indicates that the reference variable (rows) was included in the full model to explain the response variable (columns).Reference variables were excluded if they were used to calculate the response (red) or are highly correlated with the response (pink).Light availability was also excluded because it cannot directly affect N pools and fluxes (gray).Reference variables from a single soil depth were used to explain changes in that same soil depth, i.e. multiple soil depths were not included in the same model.S3.Analysis of variance summary from model selection procedures to explain Δ nitrification (0-5cm), Δ mineralization (0-5cm), Δ ammonification (5-15cm), and Δ mineralization (5-15cm) (Δ = Inv.-Ref.) using reference variables and M.v.biomass.Year is included as a random effect in all models.The first model selection procedure is to reduce the number of reference variables.The second model selection procedure is to add an interaction term with M.v.biomass to each reference variable retained from the previous selection procedure.M.v.biomass is always retained as a fixed effect.Fixed effect coefficients that significantly differ from zero are bold (alpha=.05).The significant nitrate and moisture terms for Δ nitrification (0-5cm) and Δ mineralization (0-5cm) should be considered with caution since the highest nitrate and moisture sample is very influential (see Fig. S5).

Figure S1 .
Figure S1.Each study site consisted of a reference and M.v.-invaded plot that were positioned to sample a six-meter transect across the invasion boundary.During peak growing season, August in 2012 and 2013, the three 0.25 m x 0.25 m quadrats were aligned 1 m apart and perpendicular to the invasion boundary.From each quadrat, M.v. and non-M.v.understory vegetation were harvested, plant litter was collected, and soil was sampled.In 2012, trees at each site were surveyed in within a 10 meter radius centered on the 2012 transect invasion boundary.In 2012, light availability was measured at 1m above each quadrat and averaged by plot.

Figure S4 .
Figure S4.Initial piecewise SEM constructed for Δ mineralization (5-15cm).Regression 2 is based on the final model proposed by model selection procedures (TableS3).The data do not adequately fit the SEM model (C = 18.34; df = 8; p = .02).For this reason, a causal link was subsequently added between TreeBA and Δ mineralization (Fig.2; C = 10.22;df = 6; p > .05).Values on top of arrows indicate the standardized path coefficient and solid bold arrows are significant (alpha = .05);marginal R 2 values are provided.An arrow that points to another arrow illustrates an interaction term (e.g.Fig.1, arrow D).Tree BA is shorthand for "Tree basal area".
Figure S4.Initial piecewise SEM constructed for Δ mineralization (5-15cm).Regression 2 is based on the final model proposed by model selection procedures (TableS3).The data do not adequately fit the SEM model (C = 18.34; df = 8; p = .02).For this reason, a causal link was subsequently added between TreeBA and Δ mineralization (Fig.2; C = 10.22;df = 6; p > .05).Values on top of arrows indicate the standardized path coefficient and solid bold arrows are significant (alpha = .05);marginal R 2 values are provided.An arrow that points to another arrow illustrates an interaction term (e.g.Fig.1, arrow D).Tree BA is shorthand for "Tree basal area".

Figure S5 .
Figure S5.(a, b) Nitrification (0-5cm) and (c, d) mineralization (0-5cm) differences between paired reference and invaded plots are best explained by reference soil (a, c) nitrate and (b, d) moisture (TableS3).Each point represents a site (n = 16) and year (2012 = circles, 2013 = triangles).Model fits and 95% prediction intervals are shown conditional on year (2012 = solid line, 2013 = dotted line; prediction intervals overlap in this figure).Significant relationships presented in panels a, c, and d should be considered with caution since the highest nitrate and moisture sample is very influential.

Figure S6 .
Figure S6.differences between paired reference and invaded plots are best explained by (a) an interaction between M.v.biomass and reference soil nitrate, (b) tree basal area, and (c) understory biomass (TableS3).Patterns associated with Δ mineralization at 5-15cm are analogous (Fig.3).Each point represents a site (n = 16) and year.For panel a, lighter point color represents larger soil nitrate concentrations.For panels b and c, model fits and 95% prediction intervals are shown conditional on year (2012 = solid line, 2013 = dotted line; prediction intervals overlap in this figure).
Figure S6.differences between paired reference and invaded plots are best explained by (a) an interaction between M.v.biomass and reference soil nitrate, (b) tree basal area, and (c) understory biomass (TableS3).Patterns associated with Δ mineralization at 5-15cm are analogous (Fig.3).Each point represents a site (n = 16) and year.For panel a, lighter point color represents larger soil nitrate concentrations.For panels b and c, model fits and 95% prediction intervals are shown conditional on year (2012 = solid line, 2013 = dotted line; prediction intervals overlap in this figure).

Figure S8 .
Figure S8.Relationship between tree basal area (m2), light availability (log %), and understory biomass (g/m2).For panels a, each point is a site (n = 16).For panels b-c, each point is a site and year (2012 = circles, 2013 = triangles) with gray lines connecting samples from the same site.Understory biomass increases with tree basal area (panel b, p = 0.01).All other relationships are non-significant.

Figure S9 .
Figure S9.Relationship between understory biomass (g/m2) and invader biomass (g/m2) across sites has a weak hump-shape.Each point is a site (n = 16) and year (2012 = circles, 2013 = triangles) with gray lines connecting samples from the same site.

Table S2 .
Summary of the effects of M.v.invasion on soil N pools and fluxes.Plot type (invaded or reference) is the fixed effect for which estimates, standard error, degrees of freedom and pvalues are shown.Random effects included site and year.Fixed effect coefficients that significantly differ from zero are bold (alpha=.05).

Table S4 .
Analysis of variance summary from model selection to explain M.v.biomass using reference variables.Year is a random effect.Fixed effect coefficients that significantly differ from zero are bold (alpha=.05).