Neutral‐based processes overrule niche‐based processes in shaping tropical montane orchid communities across spatial scales

Mountains harbour a global hyperdiversity of species and endemism (Enquist et al., 2019; Küper 2004; Peters 2019). The complex geological history mountains boosted the evolution many plant groups by promoting speciation through, for instance, provision new habitats, or ecological dispersal barriers to other populations same (Luebert & Weigend, 2014; Pérez-Escobar, Gottschling, 2017). replacement between sites—compositional turnover—is prevailing pattern across African (Peters 2019), European (Fontana 2020) Andean mountainous regions (Herrera-Pérez Muñoz Mazón 2021; Novillo Ojeda, 2021). However, processes governing high compositional turnover typical ecosystems remain understudied at multiple spatial scales. Patterns in can be explained niche-based processes, that is adaptation local environmental conditions, neutral-based such as limitation (Hubbell, 2001; Leibold Tuomisto 2003). Niche-based act when communities respond heterogeneity due availability niches coexistence with (Chase Leibold, In regions, variability habitat patchiness, drastic changes precipitation, topography increase (Arellano 2016; Du Trujillo contrast, become important their fluctuate random demographic stochasticity spatially limited Rosindell 2011). Species dispersion sometimes propagate successfully colonise within narrow geographical ranges regardless heterogeneity. Strong expected abiotic factors, topographic landscape fragmentation, especially dominated rare 2017; Myers 2013; Willig emerge from alike. If specialists very specific because cannot disperse long distances, both, then it reasonable expect large pool these exhibit turnover. Studies woody plants elevation gradients have shown are temperate forests (e.g. 2013) both niche- drive community structure tropical forests, probably recruitment strong Tello 2015; while vascular epiphytes seem near-neutral (Janzen 2020). These further influenced scale study anthropogenic activities, loss (Arroyo-Rodríguez 2013, plays an role how changes. Geographic distance shapes regional composition patterns (neutral-based processes). Equally, differences operate mostly scales (Karp 2012) longer steeper allow more marked sorting different preferences (Trujillo even human-modified landscapes (Du Sreekar Human modification natural also alters (Haddad Venter 2016). novel conditions generated human activities trigger homogenization pools, process which small set organisms cope flourish after alteration is, biotic (McKinney Lockwood, 1999). For transformation reduces structural connectivity causing harsh inhibit forest-specialist dispersal-limited 2017), habitat-specialists aggregation reduction (Audino Karp 2012). A better understanding required quantify influence neutral- landscapes. Empirical research now dedicated disentangling contributions using pattern-to-process approaches scales, controlling spurious environment-space correlations (Clappe 2018). We quantified roles (i.e. limitation) montane region. particular, we quantified: (i) elevations; (ii) independent combined predictors (natural anthropogenic) elevational distance) shaping focus on Colombian Andes, area highly diverse topographical hydrological (Bürgl, 1967) cause shifts biological over short distances (Hazzi 2018; Malizia 2020; Rahbek use orchids model group Andes centre orchid diversity (Pérez-Escobar 2022) one highest richness worldwide, wide variation species' range sizes widespread restricted (Calderón-Saenz, 2007; Jost, Moreno majority wind main strategy (anemochory syndrome), interspecific competition, effects herbivory pathogens negligible (Dressler, 2005; Zotz, hypothesised distinct another, type (high turnover). would driven primarily variables, study, been documented Mondragón Zuleta was located departments Cundinamarca, Boyacá, Meta Santander, spanning east west flanks eastern cordillera Colombia (Figure 1). Between January 2018 November 2019, sampled 332 plots 270 km distance, covering range, 1163–3763 m (4–28°C 879–3817 mm per year). included three types: pasture cattle (iii) paramo shrublands grasslands. All habitats undergone historical human-mediated disturbance given easy accessibility roads footpaths, information field assistants each sample location (pers. comm. landowners). belonged network protected areas (Sistema Nacional de Protegidas—SINAP). sampling design followed observed tree line along gradient S2). treeline considered upper-limit boundary establishment arboreal vegetation upper forest low-stature (Bader 2007). Below (at elevations 1163–2900 m), paired nearby pastures, above treeline, shrublands, grasslands elfin were pasturelands 1; henceforth >2900 m). line, randomly placed 1 18 (n = 150 23 fragments) keeping minimum 172 apart (range 172.3–2759 Larger had ensure broader coverage. least 30 edge roads. up 3 nearest forest, 60 away 193 193-4225 m; n 114 22 pasturelands). has density, canopy cover lower solar radiation contrast consisted open matrix Poaceae sparse trees. (andino altoandino; Etter 2021) mean cloud 82% (25%–90%; Wilson Jetz, Above 72 six habitats), high-elevation 6 195 195–15.170 High-elevation pastures 500 200 200–4305 paramos). Each plot 10 × plot. recorded 2 above-ground ground, standing trees, trunks, fallen trunks branches, vines, lianas, leaves trees herbaceous plants, palm ferns cycads understorey. Canopy might but since used sites, all probability capturing orchids. Only adult individuals recorded. Adult individuals, population, evidence developed floral reproductive structure, ramets, various stems stand prominent root systems. no flowers taken nurseries vicinity plots, later visited identification. Identification morphospecies conducted following specialised literature consultancy experts Herbarium VALLE (Palmira, Colombia). two sets uncorrelated variables S1) well-recognised drivers Walther 2002; Environmental (habitat cover) (geographical elevation). Habitat composed diameter breast height (average DBH plot) density (trees reduced single predictor principal component analysis (PCA), hereafter “Forest structure” (first PCA axis accounted 61% variation, SD 1.1043, supplemental material). Cloud (multiannual mean) extracted high-resolution map 15-year MODIS archive twice-daily observations radius sites (between 2000 Mean annual precipitation CHELSA 1980 MAP; Karger percent (Fahrig, 2013). percentage proxy calculates amount surrounding our plots. Forest metric quantifies available resource inform connected (structural connectivity). Thus, operational concept driver Banks-Leite Püttker 2015). measured 30-m resolution change (Hansen 2013), converted into binary forest/non-forest (non-forest <= 50, >50% “forest” Supporting Information). selected 1000 collate match climatic predictors, did not evaluate scale-of-effects (Jackson Fahrig, defined based coordinates (XY coordinates) (m) standardised 12.5-m ALosPalsar's Radiometric terrain (Tadono 2014). Moran's Eigenvector Maps (hereafter MEMs; Dray 2006). orthogonal vectors maximising autocorrelation (calculated coefficient), acting explicit multiscale combine fine broad (Dray First, MEMs obtained projected WGS1984 Gabriel's neighbour W standardisation (sums links number plots). Second, created including latitude coordinate point corresponding generate XY coordinate-like vector. Then, calculated neighbourhood-by-distance method globally Finally, grains whole gradient, divided ranging broadest (MEM associated eigenvalues) finest lowest eigenvalues; Menegotto This approach suits nested positive autocorrelation. intrinsic constitutes challenge studies. retained (MEMs elevation) isolation shows seeds tend land <5 mother (Acevedo Kindlmann Tremblay, 1997), resulting rapid in-situ (Pérez-Escobar, Chomicki, Furthermore, short-distance common epiphytic communities, epiphytism dominant syndrome (Zotz, 2016), dwelling (Cascante-Marín 2009; Einzmann 2017a, 2017b). ran analyses expectations: drivers) extents (whole bands), first expectation (regional local) second (broad MEMs). tested occurrence site matrices calculate Simpson dissimilarity index referring (species replacement) Sørensen (Baselga, 2010). Compositional zero (no shared sites), shared. To explore general turnover, generalised additive mixed (Wood, fitted average response variable, fixed-effect effect, binomial family. account correlations, added covariance At scale, pairwise (betapart::beta.multi). this analysis, averaged fragment 23), pastureland 22), 8). robust estimate samples (Pfeifer Olivares Kessler, least-squared models exponential class centroid detected (Moran I test 0.529, p-value 0.001; Table S1). Our variable explanatory paramo), compared against versus [pastures + paramo], [forest pasture]. established units 171 out plots) paramos) among types (forest vs. paramo, paramo). computed value dissimilarities point: within-forests, within-pastures, pasture, within-paramo, forest. linear-mixed fixed effects, autocorrelations (formula glmmTMB notation Information; Brooks, Kristensen, Mittermeier, data one-skewed few values (85.7% appear less than five S2), so subtracted multiplied −1 make zero-skewed fit zero-inflated preserving data. beta distribution family (suitable proportional bounded limits, one; Cribari-Neto Zeileis, 2010), error log link, (zero-inflated formula control effect (local scale) hump-shaped seen total abundance (E. Parra-Sanchez, unpubli. data; Figure dataset low-mid band where peaks (‘band-1’; 1150–2500 m) mid-high decreases (‘band-2’; 2501–3500 allows us decompose determine relative (environment) (distance) partitioning jointly Moran spectral randomization (MSR) method. MSR accounts (Borcard 1992; Clappe Peres-Neto 2006; Wagner Dray, Variation widely implemented statistical tool disentangle (via quantifying importance components Smith Lundholm, decouples four fractions variation: [a] unique environment, [b] environment space combined, [c] [d] non-explained (residual) variation. pure inferred if fraction larger fraction, without interpreted limitation, fraction. constrained null partitioning. Given ubiquity data, bias direct inflate component, underestimating separation structured via confronting result what under neutral dynamics relationships autocorrelated analyses: partitioned bands (low-mid band-1 band-2). performed (GD environment), (Ele environment) together Ele; except analysis). MEM's predictors. Prior analyses, Hellinger transformed (Legendre Gallagher, 2001). analysed centred adjusted R2 forward selection procedure (Peres-Neto scaled variance +/−1. fractions, (Stéphane Analyses carried R Statistical Software (v4.1.3; Core Team, packages tidyverse formatting (Wickham 2019); multivariable adespatial 2017) vegan (Oksanen 2017); imputation mice (Buuren van Buuren Groothuis-Oudshoorn, 2011), linear (Brooks, gamm4 (S. Wood Scheipl, 2020); betapart (Baselga Orme, 2012); scientific figures ggplot2 Winston, 2015) maps QGis 3.16 (QGIS.org, found 331 (277 54 morphospecies) (51%) most Dichaea morrisi (recorded Epiphytic 91.6% (303 28 terrestrial species). Orchid mainly singletons (44.2%), doubletons (18.5%), fewer records (23%) just 4% (15 species) (Table coverage optimal (coverage 0.99 overall, Information, Eleven science habitat, published descriptions forthcoming data), pastures. High elevations, extents. Considering possible pairs comparisons beta-diversity (beta 0.88). remained equally low mid decreased inflection around 2800 (GAMM; Estimate 3.14, z-value 10.72, edf 3.05, 0.02; 2). (mean, 0.79, paramos 0.90, 0.98, 3a). statistically significant (within-forest t-value 0.89, 0.38; within-forest within-pasture, 1.40, 0.18). higher (mean 0.33–1), within-pastures 0.92, 0.25–1), within-paramos 0.81, 0.40–1), within-forests 0.78, 0.25–1, 3b). show only forest-pasture (z-value −4.06, < 0.01). Model performance good overdispersion (ratio 0.80, 0.06) zero-inflation issues (zero-inflation test, ratio 0. 99, 0.98; correlation >35%). bands. Specifically, significantly (band-1, forest-pasture, −2.79, 0.01; 2.47, 4a). degree confidence (Test dispersion, 0.96, 0.78; zero-inflation, 0.99, <0.28). Likewise, (band-2; 4b), 1.85, 0.04; Information 2) within-paramo -1.04, 0.03) different. presented behaviour (dispersion 1.49, 0.22; <0.19; Geographical (proxy process) best consistently types, albeit residual (unexplained) largest (p-value 0.05; 0.18, 0.26, 0.001). form proportion 10.9%–15.5% Ele 0.3%–1.1%) Environment; GD 10.4–14.9 0.06–0.90; 5). Elevation explaining analysing (2.2% 1.4% environment). fine-scale (variance 2.4 %; MEM 20–30) broad-scale (GD, 7.4%; 1–10; se, although (band 1, 1.9 0.03 elevation; 2, 2.6 Environment (0.11%–1.39%, Across responsive major involved (15.6% 0.03% (5.82 % 0.01% environment; provide much validated (85.7%) habitat. Although several studies taxa, captures (within habitat; Wind-dispersed attain sensitive (Qian Guo, 2010; findings consistent previous epiphyte abundant (Benavides 2011; Janzen Laube Communities extremely diversity-clusters some clumped 2020