*3.2. Hypothesis Testing*

### 3.2.1. Hypothesis 1: Dense Habitats Should Have High Plant and Animal Diversity

Results were not consistent with our first hypothesis. Although high woody-plant abundances in the low plain and slope were generally associated with high plant taxonomic and phylogenetic diversities, the greatest herbivore abundances and diversities were not necessarily in these same habitats. Woody-plant density and richness was ~2–3-fold lower on the plateau compared to the slope and low plain (Table 1; Figure 2a,b). Similar patterns were found for phylogenetic diversity: the plateau had the lowest diversity and the low plain was intermediate (Figure 2c,d). In contrast, while herbivore abundance was also ~2-fold lower on the plateau than low plain, abundance on the slope was similar (Figure 2e) and richness on the slope was much lower (Figure 2f). Rarefaction revealed the greatest plant-species richness and lowest herbivore richness on the slope (Figure S2). Patterns of phylogenetic diversity also differed, with herbivore *ses*MPD being greatest on the plateau (Figure 2g) and *ses*MNTD not differing significantly across habitats (Figure 2h).

### 3.2.2. Hypothesis 2: Herbivore and Plant Diversity Should Be Positively Correlated

Contrary to our second hypothesis, there was not a significant positive correlation between the abundance and diversity of flora and fauna within and across habitats (Figure 3). In contrast, there was a significant negative correlation between the richness of flora and fauna within and between habitats (Figure 3b) and we found evidence only for overall habitat-level differences in the abundance and phylogenetic diversities of plants and animals (Figure 3a,c,d). There was an interesting contrast between herbivore phylogenetic diversity in the plateau and low plain: *ses*MPD, which is sensitive to deep patterns in the phylogeny, was greater in the plateau; *ses*MNTD, which is sensitive to variation near the tips, was greater in the low plain (Figure 3c,d). Thus, plateau sites hosted large herbivores that both represented species from disparate mammalian lineages and that were likely to include closely related pairs of species from those lineages.

**Figure 2.** Abundance and diversity of (**<sup>a</sup>**–**d**) woody plants and (**<sup>e</sup>**–**h**) large herbivores across habitats (mean ± s.e.). Letters above each bar indicate significant differences based on Tukey's HSD following ANOVAs. Abundance is the (**a**) total plant stems/site and (**e**) total herbivore RIA/site. Richness is the mean count of (**b**) woody plant and (**f**) large herbivore species. Phylogenetic diversity was measured as (**<sup>c</sup>**,**g**) *ses*MPD and (**d**,**h**) *ses*MNTD, with positive values indicating overdispersion and negative values clustering.

**Figure 3.** Relationships between herbivore (**a**) abundance, (**b**) taxonomic richness, (**c**) *ses*MPD, and (**d**) *ses*MNTD with corresponding measures for woody plants. For each pair of metrics, we constructed initial linear models with predictors including the woody-plant variable, habitat, and the plant variable × habitat interaction. We simplified models based on the subset of statistically significant variables. We report adjusted R<sup>2</sup> for the model that included a significant correlation (**b**). Horizontal lines show significant differences between habitats (**<sup>a</sup>**,**c**,**d**). Colors show habitat types.

3.2.3. Hypothesis 3: Spatial Links in the Composition of Plant and Animal Communities

Consistent with our third hypothesis, spatial turnover in woody-plant and largeherbivore communities was congruent. There was significant taxonomic dissimilarity between communities of plants and herbivores across habitats (Figure 4a,b). Taxonomic differentiation was strong across habitats, but incorporating information on phylogenetic variance eroded the signal of habitat associations (Figure 4b,d). There were significant positive correlations between local plant and herbivore community compositions, even after accounting for spatial proximity and phylogenetic variance, but these correlations were also strongest when accounting only for species composition (Figure 4e,f). About half of the tree species were significantly associated with or avoided a habitat (25/47, 53%; Figure S3a). Of these, most were negatively associated with the plateau (20/47 species, 43%) and only one had affinity for it (*Acacia drepanolobium*; Figure S3a). Eight tree species were significantly associated with the low plain, including several that were also negatively associated with the plateau (*Balanites glabra*, *Acacia gerrardii*, *A. etbaica*, *A. brevispica*, *Acokanthera* sp. 1, *Pyrostria* sp. and two *Grewia* spp.; Figure S3a). Many tree species had strong and positive affinities for the slope, especially those that also had positive affinities for the low plain, although slope sample sizes were small and results were not statistically significant (Figure S3a). In contrast to plants, nearly all large herbivore species exhibited significant habitat association or avoidance 94% (16/17; Figure S3b). Most species exhibited preferred either the plateau (9/17, 53%) or low plain (7/17, 41%), while none were associated with the slope and most avoided it (11/17, 65%; Figure S3b).

**Figure 4.** Taxonomic and phylogenetic differentiation in woody-plant and large-herbivore communities. Nonmetric multidimensional scaling (NMDS) revealed significant compositional dissimilarity in (**a**) tree PCD, (**b**) tree PDCc, (**c**) large-herbivore PCD, and (**d**) large-herbivore PCDc between habitats. Colors in (**<sup>a</sup>**–**d**) indicate habitat at each grid site. Results of perMANOVAs and stress values for corresponding NMDS plots are shown on each panel. We grouped the three slope sites with low plain sites for calculating perMANOVAs and 95% confidence ellipses. The (**e**) PCD and (**f**) PCDc of plant and animal communities were positively correlated based on Mantel and (p)artial Mantel tests; green trendlines were fit using generalized linear models.
