Taxonomic, phylogenetic and functional diversity estimation at alpha scale
We computed alpha-diversity metrics for each plot using the framework of Chao and colleagues (Chao et al., 2014), which uses Hill numbers that differentially weigh common and rare species for each of the following biodiversity facets–taxonomic, functional and phylogenetic. We used the Hill number q =0, where species are weighted equally (i.e., species richness for taxonomic diversity), and the Hill number q = 2 (equivalent to Simpson’s diversity in taxonomic analyses), where abundant species are strongly influential and rare species largely ignored. By comparing Hill numbers when q=0 versus q=2, we can detect if diversity patterns are driven by rare or common species (we also used q=1, which is equivalent to Shannon diversity, but the results were largely intermediate between these extremes; Figure S1). These metrics express diversity in multiple facets in comparable units, where taxonomic diversity quantifies the effective number of equally abundant species, phylogenetic diversity (PD) quantifies the effective total branch lengths and functional diversity (FD) quantifies the effective number of equally distinct virtual functional groups. This allows a meaningful and direct comparison across the three diversity facets (Chao et al., 2021). The framework also controls for plots with different sample sizes; here we used sample-coverage-based rarefaction (sensu Chao et al., 2021), which standardizes to a level of sample completeness. We compared taxonomic, functional and phylogenetic diversity estimation of plots using a sample completeness of 80% (i.e., sample coverage), for both analyses that included only native species and for both native and alien species. The standardization of diversity metrics via controlling for sample coverage, allow for a direct comparison among plots and diversity facets (Chao et al., 2021).
For taxonomic diversity, we used the taxonomic names (and abundances) of each species to calculate diversity metrics while controlling for differences in sample effort using coverage. For phylogenetic diversity, species abundances and species evolutionary histories (based on Faith’s metric (Faith, 1992)) are included to quantify the number of equally divergent lineages. To compute Faith’s metric, we first built the phylogenetic tree with native species (n=79) and with all species (n=142) using data of the mega-tree for vascular plants as the backbone (GBOTB.extended.tre; Jin & Qian, 2019). We resolved missing species by randomly placing them within genera or families (“scenario 2”) using the phylo.maker function in the R package ‘V.PhyloMaker’ (Jin & Qian, 2019). For functional diversity, we used four traits related to the plant’s ecological responses to precipitation. Specific leaf area (mm2/mg) because smaller leaves with higher SLA prevent water loss in dry environments (Westoby et al., 2000). Stem-specific density (g/cm3) relates to drought tolerance, as it tends to increase in dry environments (Barajas Barbosa et al., 2023). Maximum plant height (m); plants can reach higher heights in wet environments because substantial water is required to maintain their hydraulic safety (Liu et al., 2019). Leaf Nitrogen content (mg/g) relates to plant ecological strategies (i.e., fast-growing versus slow-growing) to tolerate and recover from environmental stressors, such as drought (Díaz et al., 2016). We sourced trait information from the Flora of the Hawaiian islands (Wagner and Khan et al 2023) and the TRY database (Kattge et al., 2020). In all, for the 142 species in Open Nahele, we found data on these four traits for 70% of the species (species by trait matrix; Figure S2a). We filled the 30% of missing trait values in the species by trait matrix using phylogenetic imputation (Penone et al., 2014). For this, we used the previously built phylogenetic tree (sensu Jin & Qian, 2019) including native and alien species from OpenNahele (n = 142; Figure S3). We then used a random forest algorithm using the ´missForest‘ function from the MissForest R package (Stekhoven, 2013) to predict the 30% of missing trait values. We measured the prediction error of the random forest algorithm that included phylogenetic relationships using out-of-bag error (OOB; Figure S2b). We checked for the differences between empirical and predicted trait values using Kruskal-Wallis-Test (Figure S2b). Functional diversity is computed from species pairwise distances based on the Gower distance. The measure qFD quantifies the effective number of equally distinct functional groups (or functional ‘species’) at a given threshold level termed tau. We used tau = dmean , i.e., mean distance between any two individuals randomly selected from the pooled assemblage (Chao et al., 2019). We calculated taxonomic, phylogenetic and functional diversity metrics using the ‘estimate3D’ function from the iNEXT.3D package (Chao et al., 2021).