2. 1. Study System – Temperate oak savannas and woodlands
The savannas of the Midwestern United States are dominated by oak trees, most often by bur oak (Quercus macrocarpa) with white oak (Q. alba) and black oak (Q. velutina) at lower abundances (Curtis, 1959). Oak savannas often share species with neighboring deciduous forests and tallgrass prairies which all occur in a mosaic across the landscape (Anderson, 1987). In addition, there are several unique community types within or closely related to oak savannas, based on soil moisture availability, soil properties, and overstory tree composition (Curtis, 1959). For example, on the wet end of this gradient are oak woodlands, historically dominated by swamp white oak (Q. bicolor) and bur oak (Q. macrocarpa) (Curtis, 1959), and on the dry end of this gradient are cedar glades, which are dominated by red cedar (Juniperus virginiana) (Curtis, 1959). Woody encroachment has affected all of these community types, regardless of their environmental differences (Mills, 2008; Ladwig et al., 2018). Here we use "oak savannas" to refer to oak woodlands, oak barrens, and other fire-maintained communities with sparse canopies and grassland understories.
Reduced fire frequency and intensity plus altered grazing has facilitated the encroachment of woody species into prairies and savannas (Figure 1a) (Cottam, 1949; Heisler et al., 2003; Briggs et al., 2005; Peterson et al., 2007; Stevens et al., 2016; Wilcox et al., 2018). In fire- or grazing-maintained savannas, scattered and sparse trees generate light gradients (Brudvig and Asbjornsen, 2009; Bray 1955; Leach and Givnish, 1999) that promote a rich herbaceous understory and high plant diversity due to heterogeneous fine-grained microclimatic conditions (Leach and Givnish, 1999). Shady conditions near the sparse trees also allows shade-tolerant and fire-sensitive species to establish in savannas (Mariano et al., 2018; Kreye et al., 2013). When woody encroachment occurs, the canopy of the savanna closes and microsite variation in nutrients, soil moisture, and light becomes more homogenous (Breshears, 2006), reducing plant species diversity (Anderson, 1998).
2.2. Vegetation Sampling
We use vegetation survey data from 1951 to 1954 (hereafter referred to as the 1950s) collected from oak savannas and cedar glades across southern Wisconsin (42 – 45° N, 88 – 93° W; Bray, 1960) that were surveyed as part of larger project to classify the vegetation of Wisconsin (Curtis, 1959; Waller et al., 2012). Sites surveyed in the 1950s were chosen because they retained sparse, open-grown trees and understories dominated by native grasses and forbs (Bray, 1960).
In the summer of 2014, 16 of these savanna communities were resampled following the same methods as the initial 1950s survey (Ladwig et al., 2018). At least 20 sampling points located >10 m from each other were surveyed for understory vegetation at each site. A 1 × 1 m quadrat was placed North and center of each sampling point and presence of all species occurring in the quadrat were recorded. See Ladwig et al. (2018) for additional sampling details. Taxonomic resolution was kept consistent between survey times. In the original surveys, most plants were identified to species, but some were identified to genus (e.g., Carex spp.). When there was a mismatch in nomenclature between the 1950s and 2010s, nomenclature was updated to match modern flora (Chadde, 2019). Across both survey times, a total of 261 plant taxa were identified in savanna understories.
2.3. Functional Traits
To test how functional traits were associated with plant response to woody encroachment in savannas, we measured functional traits related to persistence and dispersal for all understory forbs, grasses, and vines, excluding shrubs and tree seedlings. Categorical traits were determined from regional floras or online databases as described below. Continuous quantitative traits were measured following methods described by Pérez-Harguindeguy et al. (2013) from 10 individuals from each of three populations (30 individuals total) from remnant grasslands and savannas in central WI during the 2014 to 2019 growing seasons. Missing trait data were filled with comparably collected trait data from Wisconsin (Waller et al., 2021) and other midwestern states (Zirbel et al., 2017). Any remaining gaps were filled with identically collected data from the same species in southeastern United States longleaf pine savannas (Damschen et al., 2019; Orrock et al., PNAS, In Revision). Traits used in analyses are taxon averages of all measurements.
To characterize persistence ability, we assigned all species persistence categories based on the presence of belowground structures and life history characteristics (Miller et al., 2017). Persistence categories ranged from 1 to 5, corresponding to the least to most persistence capacity as follows: (1) annual and biennial species; (2) perennials without root storage structures or rhizomes; (3) perennials with storage structures such as tubers, taproots, corms, or bulbs; (4) species with short rhizomes; and (5) species with long rhizomes. To assign species to these persistence categories, we determined the lifespan and types of belowground tissue present from a regional flora (Chadde, 2019).
To characterize dispersal ability, we used dispersal mode category, inflorescence height (cm), and seed mass (mg). Dispersal mode was categorized as: unassisted (no specialized dispersal structures), wind dispersed (with pappi or wings), or animal dispersed (with fleshy fruits or adhesive dispersal structures) (Bullock et al., 2006) using the USDA Plant Database (USDA, NRCS. 2022; http://plants.usda.gov) and primary literature. Plant height and seed mass were measured in the field using standard methods (Pérez-Harguindeguy et al., 2013).
2.4. Statistical Analysis
To compare changes in functional traits independent of their impact on species abundance, we used a separate t-test (using the function ‘t.test’ in the stats package) for each trait to test for differences in the mean trait values of all species present between the two time periods. We also used a t-test (using the function ‘t.test’ in the stats package) to determine change in overall community abundance.
Traits used in analyses are averages of all measurements. To test whether species abundances (percentage of plots occupied) are influenced by persistence or colonization abilities, we evaluated the influence of traits and time on species abundances by running a generalized linear mixed model for each trait using ‘glmer’ in the lme4 package (Bates et al., 2015). We used the frequency of plots occupied out of the total sampled at each site as a measure of species abundance for each site. We used logit-normal binomial generalized linear mixed models (GLMM) following Jamil et al. (2013) and Miller et al. (2018) with the abundance of each species at each site as the response variable and species’ trait values and time (1950s and 2014) as fixed effect predictor variables. Seed mass, inflorescence height, and persistence rank were all included as continuous traits and dispersal mode as a categorical trait. Persistence rank was treated as a continuous instead of a categorical variable because it is a scale of relative persistence ability with accumulating persistence effects. Treating it as a categorical trait would incorrectly assume independence of each rank. The model also included independent random effects of species taxonomic identity, species-dependent random effects of time, and site- and time-dependent random effects of trait values and observation (each unique combination of species, site, and time). Specifically, the models were fit as follows:
glmer(abundance trait + time + trait:time + (1|species) + (0 + time|species) + (0 + trait|site:time) + (1|observation), family = “binomial” (link = “logit”))
The significance of each fixed effect, including interactions, was determined from asymptotic Wald tests.
Linear regressions were run between categorical traits to test for correlations. All statistical tests were conducted in R version 4.1.1 (R core team, 2021) with an alpha of 0.05.