Landscape structure shapes tree seedlings’ diversity at multiple spatial scales in a fragmented tropical rainforest

Biotic-dispersed tree seedling species are fundamental for the maintenance of the structure and function of forest patches in fragmented rainforest landscapes. Nonetheless, the effects of landscape structure and the spatial scale at which operates on seedling α- and β-diversity is unknown. Using a multi-scale approach, we assessed the relative effect of landscape composition (i.e., percentage of old-growth/secondary forest cover), configuration (i.e., aggregation/density of forest patches) and connectivity (i.e., structural and functional) on α- and β-diversity of biotic-dispersed seedlings in 16 forest patches in the Lacandona rainforest, Mexico. We assessed these effects at 13 spatial scales (from 300 to 1500 m radius, at 100 m intervals) for three α- and β-diversity orders (rare, common and dominant species). We found that patch aggregation increased species richness and reduced β-diversity of common and dominant species at similar spatial scales (500 to 600 m). Additionally, functional connectivity had a positive effect on the β-diversity of rare species in the 800 m spatial extent. These effects suggest that landscape configuration and functional connectivity sustain seedling diversity by preserving seed rain richness and the presence of large terrestrial herbivorous mammals. In contrast, the percentage of secondary forest matrix was detrimental for all α-diversity orders and the β-diversity of common and dominant species. Forthcoming conservation strategies should prevent deforestation, increase habitat amount and promote functional connectivity of forest-dependent fauna through matrix management actions.

6 114 determined the dispersal syndrome (abiotic or biotic; Table S1) of each species, based on their fruits 115 and seed morphology [44][45][46][47]. Since in rainforests biotically-dispersed tree species comprise up to 90% 116 of the seed rain [16,48], we developed the analysis using only this species guild. Plant nomenclature 117 followed the Missouri Botanical Garden database Tropicos [49].  We calculated the EC using quality values that describe the permeability of matrix covers for 152 terrestrial mammals. The permeability values were based on the assumption that overall species' 153 presence declines along a gradient of habitat loss and relates the percentage of each land-cover type 154 within the landscape matrix to its relative quality [3,27]. We ranked the relative quality of each land-155 cover type based on the suitability of vegetation structure for mammals´ feeding, movement and/or 156 habitat on a seven-point scale: 1 (water bodies, with the lowest suitability); 2 (anthropogenic cover); 3 157 (cattle pasture); 4 (arboreal crops); 5 (semiaquatic vegetation); 6 (secondary forest); and 7 (old-growth 158 forest, representing the highest suitability). To obtain a more robust and realistic representation of 159 landscape structure effects, we estimated the area-weighted mean of the EC index [74].

160
We estimated these metrics within 13 circular buffers (of a 300 to1500-m radius, at 100 m 167 tests, we did not find spatial autocorrelation between the distance of sampling sites and the diversity 168 patterns assessed (Table S2). We assessed the scale of effect of each landscape metric using linear models. We previously 173 verified the variables' normality with a Shapiro-Wilk test [77]. We fitted each diversity pattern with a 174 single landscape metric for each buffer size and assessed its predictive power with a leave-one-out 10 175 cross-validation (LOOCV). Next, we calculated the proportion of the variation that can be predicted by 176 the model using the LOOCV coefficient of determination (R 2 CV ) as follows: where y i is the diversity value for the i th patch and n is the number of patches. The 192 models with a AICc difference lower than two (ΔAICc < 2) as the best supported by the data [82].

193
Finally, we assessed the importance and the relative effect of each landscape metric measured at 194 the scale of effect on each diversity value using an information-theoretic approach and multimodel 195 inference [82]. For this, we selected a subset of models that had the 95% probability of containing the 12 218 The AI and PD metrics affected 1 β and 2 β at the 500-m buffer, whereas the latter metric influenced 0 β at 219 the 1400 m buffer (Table S4). The EC affected 0 β and 2 β in the same buffer size, whereas the scale of 220 effect of SF varied greatly between 0 β and 1 β ( Figure S3).

221
When assessing the importance and contribution of the above metrics, we found that the best-222 fitting models of α-diversity included the positive effects of AI and the negative effects of SF (Table 2).
223 The former metric was the second most important variable only for 0 α (Figure 2c), whereas SF was 224 strongly (∑w i > 0.75; Figure 2a) and significantly associated with all α-diversity orders (Figure 2c). For 225 β-diversity, we found that AI and SF metrics had negative effects for 1 β and 2 β, whereas EC was 226 negatively associated with 0 β ( Table 2). The importance of these metrics was high and significant for 227 their respective β-diversity orders (Figure 2b and 2d).  Table S5). The   As predicted, α-diversity was strongly and positively associated with habitat amount and 244 declined as SF increased in the surrounding matrix. Nonetheless, habitat amount was related to AI 14 245 rather than OGF. In addition, the AI and SF affected the three α-diversity orders in the 600-m buffer.
246 Thus, α-diversity alterations by the above metrics and scales of effect suggest that species arrival is 247 influenced by small-scale drivers that affect the local seed rain.

248
Firstly, the relationship between AI and habitat amount, as well as the observed scale of effect, 249 is not surprising. In highly disturbed landscapes, the scale of effect is expected to be smaller since

259
Although having non-significant effects, the scales of effect of OGF (300-500-m; Table S4)  The negative association between 1 β-/ 2 β-diversity and SF and AI indicates that matrix 278 composition and patch clumping have homogenizing effects on seedling assemblages through 279 abundance alterations. We suggest this homogenization is driven by the input of successional-tree 280 species at larger spatial scales and the reduction of habitat heterogeneity, seed dispersal and seedling 281 predation at smaller scales.

282
The