35Aim: We aimed to assess the contribution of marginal habitats to the tree species 36 richness of the Mata Atlântica (Atlantic Forest) biodiversity hotspot. In addition, we 37 aimed to determine which environmental factors drive the occurrence and 38 distribution of these marginal habitats. Brazil where it largely occurs, stretches for over 3,500km across equatorial, tropical 73 and subtropical latitudes, and is renowned worldwide for being one of the 35 74 biodiversity hotspots for conservation prioritisation (Myers et al., 2000). Its 75 importance is also demonstrated by its designation as one of the five primary 76 vegetation 'Domains' of Brazil (IBGE, 1993; Ab'Sáber, 2003), the others being the The prevailing land cover of these bordering Domains are semi-arid thorn woodlands distribution of rain forest species in the Atlantic Domain, which at its harshest 88 extremes give rise to distinct habitats (one for each factor), referred to as marginal 89 habitats. Therefore, the rain forest is placed by Scarano (2009) (Galindo-Leal et al., 2003; Tabarelli et al., 2004; 2005; Joly et al., 2014
203The data were originally compiled from an extensive survey of published and
217It also excludes checklists with low species richness (< 20 species), because this is 218 often due to low sampling/collecting efforts, which results in poor descriptive power.
219This study used a subset of tree inventories from the NTT database, The distribution of the sites in the ordination space yielded by NMDS (Fig. 3a waterlogged soils at positive scores (tropical riverine forests).
377The floristic composition of marginal habitats is not simply a nested subset of 378 the more species rich rain forest. The turnover component accounts for most of the 379 floristic dissimilarity of each marginal habitat in relation to rain forests (Fig. 4).
380Nestedness is higher than the turnover component in very few cases (i.e., few The forward selection procedure retained 13 environmental variables in the 390 model to explain the variation in tree species composition (Table 1). In partitioning 391 the variation explained by the retained environmental and spatial predictors, we 392 found that the environmental fraction explained 27% of the variation, 5% of which 393 was independent of spatial autocorrelation (P < 0.01). The environmental predictors
394could not account for a spatially structured variation of 12% (P < 0.01), and 61% of 395 the variation remained unexplained (see discussion for more details).
396The harshest extremes of the retained environmental variables (Table 1) rock outcrops (including campos rupestres) from all others vegetation types (Fig. 3a).
404Within the rock outcrop habitat, the frequency of frost was associated with the forests and tropical riverine forests (Figs. 2b and 3b). At the harshest extreme of the 414 drought-stress gradient (Fig. 3b) Fig. S1). Because the overall floristic dissimilarity between cloud forests and rain
430forests was relatively low (Fig. 3), we assessed the rates of endemism con...