Selective manipulation of a non-dominant plant and its herbivores affects an old-field plant community

Abstract

Competition and herbivory can interact to influence the recovery of plant communities from disturbance. Previous attention has focused on the roles of dominant plant species in structuring plant communities, leaving the roles of subordinate species often overlooked. In this study, we examined how manipulating the density of a subordinate plant species, Solanum carolinense, and its insect herbivores influenced an old-field plant community in northern Florida following a disturbance. Five years following the disturbance, the initial densities of S. carolinense planted at the start of the experiment negatively influenced total plant cover and species diversity, and the cover of some grasses (e.g., Paspalum urvillei) and forbs (e.g., Rubus trivalis). Selectively removing herbivores from S. carolinense increased S. carolinense abundance (both stem densities and cover), increased the total cover of plants in the surrounding plant community, and affected plant community composition. Some plant species increased (e.g., Digitaria ciliaris, Solidago altissima) and others decreased (e.g., Paspalum notatum, Cynodon dactylon) in cover in response to herbivore removal. Herbivore effects on plant community metrics did not depend on S. carolinense density (no significant herbivory by density interaction), suggesting that even at low densities, a reduction of S. carolinense herbivores can influence the rest of the plant community. The recovery of the plant community was context dependent, depending on site- and plot-level differences in underlying environmental conditions and pre-disturbance plant community composition. We demonstrate that the density of and herbivory on a single subordinate plant species can affect the structure of an entire plant community.

Appendices

Appendix 1

See Fig. 4.

Fig. 4

For logistical reasons, plot size varied with density (Spearman r _s = −0.948) but to avoid completely confounding density with plot area, we used two plot sizes for each density treatment. Although this does present the possibility of spurious correlations with density and plot size, there was enough variation between the two variables to decouple their effects on plant responses. However, to ensure that spurious correlations were not present, we re-ran GLMs (see “Data analysis” in Methods), replacing the significant variable (e.g., density) with the other correlated variable (e.g., plot size) and compared model results and AICc values. If the replacement did not yield the same results and the other variable did not appear significant, we conclude that it was not a spurious correlation and that the effects are real. In all cases, the density and area effects represented in the paper were real

Appendix 2

Index of prior plant composition.

Plant composition prior to disturbance was estimated from high-resolution digital orthographic quarter quad imagery acquired in March 2004, 3 years before the start of the experiment. Different plant species, but also variation in soil and vegetation water content, can reflect light of different intensities across a range of the electromagnetic (EM) spectrum, which forms a basis for all land cover classifications (e.g., Jin et al. 2013). Accordingly, we assume that the digital number values corresponding to each 1 m² pixel represents the average spectral radiance of the individual plants within that area, and therefore the composition of plant species. Moreover, we assume underlying environmental variation potentially driving differences in prior plant cover (e.g., soil type or hydrological properties) are incorporated within radiance values. We included the three bands from the image (green, blue, and infrared) in a principal components analysis (PCA) in order to generate an image that represents spectral radiance from multiple ranges of the EM spectrum on a single band. Ordination techniques such as this can be used to characterize vegetation communities in a continuous fashion rather than forming discrete habitat classes (Schmidtlein et al. 2007). We overlaid the plot boundaries on the PCA image product and, for each plot, calculated the mean of the first-axis PCA scores (Fig. 5). The first axis explained 71.6 % of the variation in the original three-band image. The mean for each plot was used in subsequent analyses as an index of prior plant cover.

See Fig. 5.

Fig. 5

PCA images illustrating differences in prior plant cover among plots in the west (a) and east (b) sites. Plot positions are outlined in white and ranged in sizes from 1.6 to 98.4 m². The five initial planting densities of S. carolinense within each plot were 0.65, 2.77, 11.11, 22.68, and 30.86 stems/m². This figure depicts all three axes as a false color composite with PCA axis 1 (71.6 %) coded as red, PCA axis 2 (25.7 %) coded as blue, and PCA axis 3 (2.7 %) coded as green, but only the mean of the first-axis scores corresponding to areas within each plot was used as a covariate in our analyses “prior plant cover”

Appendix 3

See Table 2.

Table 2 The effects of S. carolinense insecticidal spray and initial planting densities on various grass (A) and forb (B) species in the surrounding plant community in 2011. Numbers in parentheses are the mean cover of each plant species

Appendix 4

See Table 3.

Table 3 The effects of S. carolinense density, insecticide spray,  and site on leaf damage to Rubus trivalis (A) and Solidago altissima (B)