Temperate savannas and grasslands maintained by frequent, low-intensity disturbances such as fire contain among the most species-rich plant communities in the world. Precisely how these disturbances maintain such high fine-scale diversity is poorly understood. This study examined the effects of the frequency of simulated fire (clipping combined with litter removal) and the relative importance of recruitment and survival on species diversity and trait and species composition at each of two pine savannas in southeastern Mississippi (USA) that had not been recently burned. Ten 2 × 2 m plots at each site were clipped/cleared annually from 2014 to 2019 and again in spring 2021 (annual frequency). The other 10 clipping plots were not clipped from 2018 to 2020 (reduced frequency). Vegetation in small subplots in annual frequency and reduced frequency plots was compared in August 2021 to test the effects of a short period without clipping on diversity and composition. To test the relative importance of recruitment and survival on diversity and composition, four 0.25 × 0.25 m quarter plots were established within each of 10 annual-frequency plots per site following a clipping treatment in fall 2019 and assigned a 2 × 2 factorial arrangement of transplantation of sods from long-unburned areas and herbicide application. Reducing the frequency of clipping reduced plant diversity and altered composition at both sites. A comparison of diversity and trait composition responses to transplant and herbicide treatments revealed how recruitment and survival combined to affect species diversity. Partial or complete recovery of diversity following clipping and litter removal at both sites was driven by rapid increases in short-lived, resilient species that show fire-stimulated emergence from a seed bank and the persistence of long-lived species capable of surviving the prolonged period without fire or clipping. Species with reduced resilience and persistence were more likely to be lost in the reduced frequency treatment. Results are consistent with a model of short-term coexistence of maximum species diversity maintained by the most frequent fire regimes fuels will permit.

Experiment 1. Effects of Clipping History and Herbicide-Generated Mortality on Fine-Scale Plant Diversity and Composition. To test the relative importance of recruitment and survival in explaining species diversity responses to frequent clipping, I examined the responses of diversity, composition, and functional traits of all identified species from 2020 to 2021 to transplantation of sods from long-unburned and unclipped areas into annual frequency plots and to herbicide application within these plots (Fig. 2). I established a 0.6 × 0.6 m subplot within the center of each of the 10 2 × 2 m Annual Frequency plots at each site in October 2019 (Fig. 2). I then subdivided each subplot into 4 equal (0.25 × 0.25 m) quarter plots and randomly assigned a 2 × 2 factorial arrangement of the transplantation and herbicide treatments to the quarter plots in each subplot (Fig. 2). I then transplanted two 0.25 × 0.25 m vegetated sods excavated from stratified random locations of portions of the savanna that had not been recently burned or clipped to the quarter plots randomly chosen to receive the transplant treatment. I chose sods in a stratified random fashion so as to avoid splitting bunchgrass clumps in half. Because many clumps of A. beyrichiana at Grand Bay were large enough to encompass most of a sod, and to avoid damage to these plants, I took most sods from locations that either were directly adjacent to an A. beyrichiana clump or that included a small (<~ 10 cm in diameter) clump of the species. This procedure resulted in an underestimate of A. beyrichiana within unclipped areas of the site. I therefore excluded A. beyrichiana from statistical analyses of the transplant treatment effect (see Data Analysis of Experimental Results). I established two transplant control quarter plots by trenching (once) around the periphery of the intact (non-transplanted) quarter plots (Fig. 2).  After transplanting the sods, I clipped all vegetation within the 2 × 2 m plot to the soil surface and removed all clippings and surface litter, as in previous years. One week later, as vegetation was beginning to resprout, I applied the herbicide treatment to a transplanted sod and a non-transplanted sod by carefully applying glyphosate herbicide (Roundup^®) to all vegetation with live, aboveground green parts.

Vegetation Sampling for Experiment 1. I monitored the vegetation responses to the quarter-plot treatments by determining presence-absence of all identifiable vascular plant species, excluding woody plants taller than 1 m, within each of four (0.125 × 0.125 m) quadrants of each 0.25 × 0.25 m quarter plot (Fig. 2). I censused vegetation in early April 2020, late May 2020, October 2020, early May 2021 (before implementation of the clipping treatment), and August 2021. I pooled the April and May 2020 censuses of spring-flowering species with the October 2020 census of summer and fall-flowering species to provide an estimate of composition in 2020. I pooled the May 2021 census of the spring-flowering species and August 2021 census of late summer- and fall-flowering species to obtain an estimate of composition in 2021. Nomenclature followed Weakley (2020) and when necessary the USDA Plants Database.

            To determine whether species richness was more negatively associated with the amount flammable fuels than with green vegetation, I quantified fine fuels, green vegetation, bare ground and species richness in May 2021 in the Annual Frequency plots. This census was conducted before the May 2021 clipping treatment and thus 19 months since the last clipping treatment. Quantification of fuels, green vegetation, and bare ground entailed taking an overhead digital photograph of each quarter plot, superimposing a 3 × 3 (9-cell) grid on each photograph, and then counting the number of cells in which the majority of the area of each cell was occupied by either brown vegetation and/or surface litter (hereafter, flammable fuel), green vegetation, or bare ground. I then examined the relationship between the species richness and the relative amounts of flammable fuel, green vegetation, bare ground, and the transplant and herbicide treatments in May 2021, before the May 2021 clipping treatment was implemented.  

Experiment 2. Effects of Reduced vs. Annual and Increased Frequency on Fine-Scale Plant Diversity and Composition. I assessed the effect of reducing the frequency of simulated fire from once every year (annual frequency) to once at the beginning and end of a four-year period (reduced frequency) on plant species diversity and trait and species composition in 2021. I established a 0.25 × 0.25 m small subplot within each of the 10 reduced frequency plots at each site in May 2021 (Fig. 3). To determine the effect of reduced frequency on plant diversity and composition, I compared the August 2021 composition of the non-treated quarter plots in the annual frequency plots to August 2021 composition in the small subplots in the reduced frequency plots (Fig. 3). To investigate how diversity and composition differed between areas that experienced a recent reduction in clipping to those that experienced a recent increase in the frequency of clipping, I compared the composition of subplots in the reduced frequency plots to those quarter plots in the annual frequency plots that had not received herbicide but had been transplanted from areas that had not been clipped or burned since at least 2010 (Grand Bay) or since 2012 (Sandy Creek; Fig. 3).

Functional Traits and their Analysis. To determine the treatment effects on functional traits, I extended the analysis of herbaceous species used in Brewer and Zee (2021) by including woody plants and species encountered in the current study that were not encountered in that study. I described all vascular plant species encountered in plots and that I could identify with respect to 25 traits (Appendix 2). Of particular interest were traits directly related to size and life history (e.g., height, root length, presence of belowground perennating organs, perenniality, presence of seed bank) and that were indicative of fire-mediated phenotypic plasticity (fire-stimulated flowering, fire-stimulated emergence, vegetative dormancy in years without fire). I described both categorical (e.g., photosynthetic pathway) and quantitative traits (e.g., leaf moisture content). I also measured traits commonly used in general functional trait analyses, including specific leaf mass (SLM), leaf moisture content, flowering season, photosynthetic pathway, and presence of nitrogen-fixing symbionts. I also measured traits that were common in wet and/or wet, nutrient poor habitats (root porosity, a proxy for aerenchyma tissue production, and carnivory; Brewer et al. 2011).

I measured quantitative traits that were continuous (e.g., specific leaf mass, leaf moisture, lateness of flowering (spore-producing) season) as well as those that were classified into two binary categories: greater than the median value or less than or equal to the median value of the species pool (e.g., maximum root length, rhizome thickness, rhizome length, root porosity). For some binary variables, there were inadequate data to classify some species. For example, for the trait “fire-stimulated emergence”, evidence for or against fire-stimulated emergence was lacking for some species. I considered responses for those species as missing.

I used the functcomp function in R (package FD) to calculate community-weighted trait means (CWM, Lavorel et al. 2008, hereafter CW trait means) using all identified species encountered in all plots for Experiments 1 and 2. For a continuous trait (e.g., lateness of flowering, specific leaf mass), the CW trait mean was simply the mean trait value of all species present in the sample (after excluding species with no measurement), weighted by their relative abundances. For binary variables, I specified CWM.type = “all”, which returned the summed relative abundances of all species in each class (excluding those that could not be scored) divided by the summed relative abundances of all species. This approach resulted in there being a mean relative abundance for each binary category (e.g., one for “carnivorous” and one for “not carnivorous”). In many cases, the CW means for the two categories had the same absolute value but differed in sign. In cases in which data were lacking for some species (e.g., fire-stimulated emergence), CW means could differ in both sign and absolute value.

Data Analysis of Experimental Results. For Experiment 1, I analyzed the effects of herbicide (herbicide, no herbicide), transplant (transplanted, not transplanted), year, site, and plot on species richness and Shannon species diversity with linear mixed models (function lmer in the lmerTest package of R; version 3.2.4). For the Sandy Creek site, I treated herbicide, transplant, year, and their interactions as fixed effects and plot as a random effect. Because of the unavoidable underestimation of A. beyrichiana abundance in transplanted sods at Grand Bay, I ran separate analyses of the herbicide and transplant effects on species richness and Shannon diversity. I conducted analyses of the herbicide effect and its interaction with year on non-transplanted quarter plots and included A. beyrichiana in the analyses. In contrast, I conducted analyses of the transplant effect and its interaction with year on quarter plots that were not treated with herbicide and excluded A. beyrichiana from the analyses. Aristida beyrichiana was absent from all plots at Sandy Creek. To increase statistical power of the tests of the transplant effect on richness and Shannon diversity, I combined data from no-herbicide quarter plots from both sites (excluding A. beyrichiana) and tested the main between-subjects effect of transplant for both years and its interactions with site and year.

I analyzed treatment effects on species and functional trait composition (based on CW trait means) for each site separately using two integrated approaches. One approach involved permutation multivariate analysis of variance (permanova) (Anderson 2001) using the adonis2 function in the vegan package of R. I used the method “bray” to conduct the analysis on Bray-Curtis distance matrices derived from quadrant sums (hereafter, abundance). To examine which species were most important in distinguishing composition between treatments, the second approach involved running a distance-based redundancy analysis and interpreting species or community-weighted mean trait scores (dbRDA; function capscale in the vegan package of R; McArdle and Anderson 2001). I analyzed functional trait composition in much the same way. As with the analysis of species composition, I used scores associated with CW traits to determine which traits were most important in distinguishing the treatments and to which treatment group each trait was most strongly associated.

For the Sandy Creek data, I examined changes in composition as a function of the transplant treatment by conducting a repeated measures permanova of the within-subjects effect of year and its interaction with transplant. This test was done without including the herbicide treatment in the model due to its overwhelming effect on mortality in both transplanted and non-transplanted quarter plots. In a separate analysis, I examined changes in composition as a function of the herbicide treatment and its interaction with transplant by conducting a repeated measures permanova of the within-subjects of year and its interactions with herbicide, transplant, and herbicide × transplant (disregarding the year × transplant test). For both analyses, I also included a grouping variable, which accounted for all the between-subjects variation and removed its associated degrees of freedom from the residual error term (see Appendix 3 to see the code for the adonis2 model statements). To isolate the effect of each treatment on compositional change and associated species or trait scores, I ran a separate partial dbRDA for each treatment × year interaction, wherein the test of each was conditioned on the grouping variable and year effects and the remaining treatment × year interactions in the model (see Appendix 3 to see the code for the capscale model statements). To generate the appropriate model statements, I converted each of the treatment interactions with year to a main effect (Appendix 4). In addition to running a permanova to determine the effects of the treatments on changes in composition, I ran separate permanovas on the 2020 abundance data (pooled April, May, October data) and the 2021 data (pooled May, August data). To examine which species and traits were most important in driving the treatment effects within each year, specifically, the herbicide, transplant, and herbicide × transplant effects, I ran a separate partial dbRDA for each of these effects and obtained species or CW trait scores. As with the partial dbRDAs of the treatment × year interactions, I isolated the main effect of each treatment on composition in each year conditioned on plot, in which the treatment was nested, and the remaining treatment effect and the interaction. To contrast the effects of herbicide and the transplant treatments on composition, I also used permanova and dbRDA to examine differences between the herbicide-only treatment and the transplant-only treatments in each year and for the change in composition between years.

For the Grand Bay data, I ran repeated measures and single-year permanovas and dbRDAs separately for the herbicide and transplant effects. As with the analyses of species richness and diversity, I analyzed the effects of herbicide on non-transplanted quarter plots and the effects of the transplant treatment on quarter plots that did not receive the herbicide treatment. To contrast the effects of the transplant and herbicide treatments on composition, I also used permanova and dbRDA to examine differences between the herbicide-only treatment and the transplant-only treatments in each year and for the change in composition between years. Because of the problem of not adequately sampling A. beyrichiana in transplanted sods at Grand Bay, I removed this species from all analyses involving transplanted sods.

I ran between-subjects permanovas of treatment effects on composition averaged over the two years when the treatment × year interactions were not significant. Between-subjects tests were based on principal coordinates analysis axis scores rather than on abundances (Anderson et al. 2008), which precluded direct identification of which species or functional traits drove compositional differences associated with the treatments averaged over the two years (Appendix 5 for the R code used to conduct the between-subjects tests).

I used linear mixed models to test the relationship between species richness and the fraction of flammable fuels, green vegetation, and bare ground and the transplant and herbicide treatments in May 2021. Before running the linear mixed models, I ran principal components analysis (using princomp in R) to quantify differences among the quarter plots with respect to flammable fuel, green vegetation, and bare ground. Two principal components were extracted. The first principal component, which explained 66% of the variation in cover among quarter plots was highly negatively correlated with flammable fuel (r_{brown vegetation} = -0.989), strongly positively correlated with bare ground (r = 0.773), and modestly positively correlated with green vegetation (r = 0.426). The second principal component, which explained 34% of the variation in cover among quarter plots was highly negatively correlated with green vegetation (r_{green vegetation} = -0.905), positively correlated with bare ground (r = 0.634) and uncorrelated with flammable fuel (r = 0.142). Hence, the first principal component was considered a flammable fuel/bare ground axis and the second principal component was considered a green vegetation/bare ground axis. I then fit species richness to a linear mixed model that included transplant, herbicide, their interaction, and the two principal components as fixed effects, and site and plot as random effects, with plot nested within site.

For Experiment 2, I analyzed the effect of the frequency treatment (annual, reduced) on species richness and Shannon species diversity for both sites together, treating site as either a random effect (using linear mixed models) or as a fixed effect. When I analyzed site as a fixed effect, I used linear models (function lm package of R) to determine if there was a significant main effect of site and a site × frequency interaction. I conducted two separate treatment comparisons for each site. For one comparison (Effect of a Recent Reduction in Frequency), I compared the control quarter plots in the annual frequency plots to the subplots in the reduced frequency plots. For the second analysis (Effect of a Recent Increase in Frequency), I compared the transplanted quarter plots that were not treated with herbicide in the annual frequency plots to the subplots in the reduced frequency plots. I analyzed the effect of the frequency treatment on composition both together and separately for each site using permanova and dbRDA. When I examined both sites together, I included site and the site × frequency interaction in the permanova model, and the partial dbRDA examined the effect of the frequency treatment conditioned on site and the site × frequency interaction.

All files can be opened using software capable of reading an MS-DOS comma separated values or plain text files.