Despite centuries of investigation, we still lack a unified explanation for the global variability in species diversity. In particular, our explanations for the large-scale diversity patterns fail to account for the enormous aggregation of species within tropical mountains. At the regional scale, speciation and persistence operating in close geographic proximity with each other may explain diversity patterns in range-restricted species. At the local scale, the partitioning of species’ functional niches may facilitate the maintenance of diverse ecological assemblages. Both of these hypotheses apply to the fine spatial scale, and have therefore been challenging to support by empirical data. For combating these challenges, the hummingbird family (Trochilidae) is suitable for several reasons. The combination of high species richness together with well-resolved information on the species’ geographical distributions and phylogenetic relationships makes the taxon suitable for investigating biogeographical hypotheses at the large spatial scale. In addition, well-sampled data on the hummingbirds’ interactions with flowering plants makes the taxon ideal for investigating the biogeographical variability in resource specialization. Using that information, this thesis addresses questions regarding variability in species diversity from regional to local scale. Major research aims include: (i) investigating the hypothesis that endemism hotspots are generated by early-divergent species aggregating with recently-derived species (Chapter 1); (ii) to disentangle the different ecological explanations for geographical variability in resource specialization (Chapter II-V); (iii) to investigate the extent to which the hummingbirds’ interactions with plants could have implications for their vulnerability to future climate changes (Chapter VI). Each chapter represent distinct research projects formatted as manuscripts. Chapter I investigates the buildup of Andean endemism hotspots by hummingbirds that are either early-divergent or recently-derived in the phylogeny. The results show that hotspots of early-divergent endemics to be spatially incongruent with recently-derived endemics. Recently-derived endemics were dispersed in low densities along the Andean treeline whereas early-divergent endemics had more aggregated distributions in cloud forest localities. Chapter II is the product of seven months of fieldwork in the Ecuadorian east Andes. Here, I noticed that the lowlands and highlands are overrepresented by distinct hummingbird bill types. Hummingbirds with exceptionally curved bills were largely restricted to the lowlands whereas hummingbirds with exceptionally long and straight bills were restricted to the highlands. Around 2000 m.a.s.l there is transition zone where neither curved-billed nor long-straight-billed specialists were present. By combining this observation with collection of three interaction networks, I proposed that it is the distribution of hummingbirds with specialized bill morphologies, which ultimately explains the spatial variation in trait matching with plants. Chapter III extends upon the previous in suggesting morphological matching between hummingbird and plant to have implications for the geographic variability in resource partitioning. The study is based on dataset consisting of 24 hummingbird-plant networks, collected with additional information on the species morphologies, phenologies and relative abundances. This information allows hummingbird-plant trait matching to be disentangled from phenological overlap and variability in abundances in constraining pairwise interactions. I identified morphological matching as a candidate explanation for the geographical variability in resource specialization. In addition, I found morphological matching and phenological overlap to be more efficient than abundance at describing species’ interaction frequencies. The ability of morphological matching to describe interaction frequencies tended to increase towards low latitudes and towards areas with low temperature seasonality. Chapter IV uses a database on 74 hummingbird-plant networks across the Americas to document a correlation between hummingbird functional diversity and two measures of network structure; both are commonly used as measures of resource specialization. Chapter V investigates the causal association between local hummingbird abundance and resource generalization. The results suggested that hummingbirds are generalized within networks because they are abundant rather than abundant because are generalized. This interpretation is based on observation that there are generally few abundant specialists whereas generalists can by both abundant and rare. Chapter VI approaches the effect of future climate changes on hummingbird-plant networks. By causing turnover in composition of community, it is expected that climate changes may disrupt the network of biotic interactions and potentially trigger local coextinctions. In the study, I simulated climate-driven extinction, coextinction and colonization within 74 hummingbird-plant networks. Particularly the simulated rate of coextinctions showed remarked variabilities between New World biogeographical regions. Coextinctions were less frequent in the Andes than in the Brazilian lowlands and Atlantic Mountains. Meanwhile, the Andean networks had large pools of colonists, which could potentially counteract species loss. The variability in simulated climate-driven extinction, coextinction and colonization was also significant between hummingbird subfamilies. Notably, the subfamilies most vulnerable to coextinctions were not necessarily vulnerable to climate-driven extinctions. Taken together, I found early-divergent and recently-derived endemics did not aggregate at the regional scale as previously anticipated. As an explanation for the geographic variability in resource specialization, the results pointed to the tendency for hummingbirds to interact with subsets of flowers that are morphologically matching their bills. Such morphological matching implied high diversity of hummingbird morphologies, which largely generates by a few species with exceptionally long or curved bills. Finally, I found significant variability in the simulated coextinctions, both at the geographical and the taxonomical scale. With this result, I speculate that biotic interactions may have implications for understanding species vulnerability to future climate changes.