1. The coexistence of closely related species is key to understanding the nature of biodiversity hotspots where in situ diversification has yielded rich communities of close relatives. Limiting similarity predicts that co-occurring species are differentiated in their niches; identifying the axes of differentiation in sympatric close relatives can thus help reveal the eco-evolutionary dynamics of community assembly. For flowering plants, these axes may be temporal (related to reproductive phenology) or morphological (related to functional traits).

2. We collected fine-scale data on abundance, morphology, and phenology over a flowering season for 34 species of Rhododendron (Ericaceae) spanning a 2700 m elevation gradient in a nature reserve in the eastern Hengduan Mountains, China. We used null models to test for patterns of clustering versus overdispersion in species’ abundances, phylogenetic relatedness, and functional traits across sites, and applied joint distribution models to examine the correlates of pairwise associations.

3. We found that species tended to be spatially aggregated, indicating that communities are not strongly structured by competitive exclusion. At higher elevation sites, species tended to be vegetatively more similar (clustered) and closely related. Environmental variables, including climate and topography, were strong predictors of species' ranges. No evidence of niche differentiation was detected along spatial or morphological (functional) axes, but along the temporal axis, the phenology of co-occurring species showed significant divergence, and was less phylogenetically conserved compared to morphological traits.

4. Synthesis. Local communities of Rhododendron in its center of diversity are structured by environmental filtering and the effects of limiting similarity. Evidence for the latter is apparent in the pervasive phenological divergence of co-occurring species, likely driven by reproductive interference from shared pollinators. The evolutionary lability of flowering time appears to render it the quickest path to coexistence for recently diverged species that experience secondary contact in this biodiversity hotspot.

Between April 25 and June 28, 2019, we censused Rhododendron species in the Gongga Mountain National Nature Reserve (29°10’–30°10’N, 101°10’–102°10’E, an area ca. 1×10⁴ km²) in western Sichuan province, China. In several major valleys surrounding Mt. Gongga, we surveyed 128 sites spaced at intervals of approximately 100 m change in elevation, or distances > 1 km in the absence of elevation change. The sites spanned elevations from 1790--4580 m, with habitats ranging from deciduous and mixed forests to alpine shrublands and meadows. We conducted four censuses, each involving a circuit of visits to all 128 sites, at intervals of 10--17 days over the fieldwork period. We estimated the population densities of Rhododendron species using variable area transects (VAT), an efficient and robust method suitable for dense populations used in previous studies of Rhododendron. For each species at a site, we measured population density once by setting up two transects from different random starting points within the boundaries of their occurrence, each with a fixed width (ω = 1.5 m). The transect length (x) was recorded as the distance to the fourth individual encountered (r = 4). Density was estimated using the formula (nr - 1)/(ω ∑xi). We translated densities to abundances by multiplying by a constant factor of 100 m². We then constructed a species abundance-by-site matrix and a corresponding binary presence-absence matrix. At each visit, we recorded the phenological phase of each species by visual estimation of the proportion of flowering inflorescences out of the total inflorescences across 10--20 individuals. These values were later normalized by the peak value observed for the species over the duration of the census. To assess functional variation related to growth, stress tolerance, and pollination, we measured three vegetative traits and six floral traits. Plant height and leaf area are indicators of light acquisition ability and correlate to overall plant biomass; specific leaf area (SLA) is a measure of photosynthetic capacity and leaf longevity, commonly used to quantify species response to environmental resources. We collected mature leaves from three individuals per species per site in sealed plastic bags to maintain freshness, and later the same day scanned them on a flatbed scanner (Epson Perfection V19) and dried them in silica gel. We measured leaf area from the scans using ImageJ, and measured leaf mass by an electronic analytical scale after oven-drying at 70°C for 24 hours. Flower size and shape are are pollination-related traits that influence the potential for reproductive interference via heterospecific pollen transfer. We measured corolla tube length and width, corolla lobe length and width, stamen length, and style length on fully-opened flowers of three individuals per species chosen randomly from representative sites (86 in total, ranging 1--10 sites per species, median = 4).

All data are in comma-separated values (csv) files, and can be easily read and processed in R.