Pollination is an important ecological process for plant reproduction. Understanding the differences in plant–pollinator interactions and pollinator importance across spatial scales is vital to determine the responses of these interactions to global changes. Continental and oceanic island systems provide us with an ideal model to examine the variation in plant–pollinator interactions. Here, we compared the differences in species composition, plant–pollinator network structure, and pollinator importance in communities between a continental island (Wanshan Island) and an oceanic island (Yongxing Island) in China. The results reveal highly dissimilar species composition between continental and oceanic islands that caused highly different plant–pollinator network structures. In particular, the oceanic island networks had higher network connectance, nestedness, and specialization than the continental island networks. For plants co-occurring on both islands, pollinator species richness and flower visitation rate were higher on the continental island than on the oceanic island. Plant niche overlap was higher on the oceanic island than on the continental island, while pollinator niche overlap was higher on the continental island than on the oceanic island in both entire network and co-occurring plant species subnetwork. Hymenoptera was the most important pollinator group in the continental island community, while Apidae was the most important in the oceanic island community. The results imply that oceanic island communities may be less vulnerable to disturbance, such as habitat alteration or destruction, than continental island communities, and provide implication insights into biodiversity conservation for pollinators on islands.

Study sites

This study was conducted on Wanshan Island (continental island; 21°56′N, 113°44′E) and Yongxing Island (oceanic island; 16°49′N, 112°20′E). Wanshan Island belongs to the Wanshan archipelago (Zhuhai, Guangdong province, China) and was separated from the mainland due to the rising sea level during the Holocene (Wang, 2008). The distance between Wanshan Island and the city of Zhuhai is about 40 km. Yongxing Island belongs to the Xisha archipelago (Sansha, Hainan province, China), a group of coral islands (Paracel Islands) in the South China Sea that was formed in the Late Tertiary (Gong et al., 1996). The Yongxing Island is developed on coral reefs, and the distance between the archipelago and the southeast of Hainan Island of China is around 350 km. The areas of Wanshan and Yongxing islands are 8.2 km2 and 2.6 km2, respectively. Our field observations of plant–pollinator interactions were conducted from July 1 to August 30, 2018, on Yongxing Island and from July 1 to August 30, 2019, on Wanshan Island.

Plant–pollinator interaction data collection

To maximize the possibility of recording different pollinators of each plant species in the community, the focal plant approach was used to collect plant–pollinator interaction data on the two islands. We observed all plant species that were in full bloom in both island communities. For each plant species, pollinators were observed and recorded on sunny days without wind between 08:30 and 17:00 h. To minimize interference with pollinator behavior during observation, observers were located at a distance of about 1 m from the focal plant species. We recorded flower visits by pollinators during 15-min periods. For each focal plant species, we recorded the number of observed flowers or flower heads, all pollinator species that visited flowers, and the number of times flowers were visited by each pollinator species. We recorded those flower visitors whose bodies touched the reproductive organs for more than one second. These visits were defined to have occurred effectively, and these visitors were classified as potential pollinators (hereafter, named pollinators). For each plant species, the flower visitation rate was calculated as the number of visits by each pollinator species during a census, standardized by the number of flowers observed and the total number of censuses. Each focal plant species in the community was observed for two consecutive days. Field counts for morphologically identifiable pollinators (e.g., Apis cerana) were recorded based on prior experience. Pollinators that could not be morphologically identified to the species level in the field were first noted to the family level. These pollinators were caught by a sweep net and then transferred to centrifuge tubes containing ethanol for further identification by entomologists. All captured insect specimens were deposited at the South China Botanical Garden, Chinese Academy of Sciences.

Composition of pollinator species between the two islands

To investigate the differences in pollinator composition between Wanshan and Yongxing islands, we calculated and plotted the proportional abundance of pollination functional groups recorded across our sampling times. The pollinator functional group includes Apidae, Syrphidae, non-Apidae Hymenoptera, non-Syrphidae Diptera, Lepidoptera, Coleoptera, and Passeriformes (hereafter, Hymenoptera and Diptera represent all the Hymenoptera and Diptera, respectively). To investigate the variation in pollinator assemblages on the two islands, we first calculated and plotted the proportional abundance of pollinator groups of seven plant species that grow on both islands. To visualize the distances of pollinator assemblages visiting the same seven plant species on the two islands, we performed non-metric multidimensional scaling ordination using the vegan package in R v.4.1.3 (R Core Team, 2022). The Morisita–Horn index was chosen because it is not sensitive to species richness and sample size (Chao et al., 2006). To do so, an interaction matrix of the seven co-occurring plant species was constructed for each island, with rows representing the plant species, columns representing the pollinator species, and cell values indicating the interaction frequency between the plant and pollinator species. To statistically test if the pollinator assemblage of a co-occurring plant species was significantly different between the two islands, analysis of similarities (ANOSIM) was performed using the “anosim” function in the vegan package in R and corroborated by 9,999 permutations. The ANOSIM test statistic (global R) is a comparative measure of average ranking within and between a priori-defined groups. In this study, the dissimilarities of the pollinator assemblage of a co-occurring plant species were ranked, and then the mean ranked dissimilarities of the two islands were compared. Global R values vary from 0 to 1; values close to 1 indicate that replicates within a group are more similar to each other than to any replicates from different groups, and 0 implies no segregation into groups. The significance of the global R is determined by permuting the membership of objects in the groups.

Plant–pollinator network structures

We constructed two quantitative plant–pollinator networks from the interaction data, one for Wanshan and one for Yongxing island communities, using the frequency of visitation of the pollinator species to each plant species as a surrogate of interaction strength between plants and pollinators. Since the value of pollinator visitation frequency is decimal, not an integer, which is not suitable for the bootstrapped analysis, we standardized the flower visitation frequency by multiplying by 1000. We compared the values of several metrics of network structure between networks with and without multiplication by 1000, and these values are the same. We visualized the two plant–pollinator interaction networks in both island communities using the “plotweb” function in the R bipartite package. The plots included the weighted interaction frequency to show interactions proportional to the number of pollinator visits. Based on these two interaction matrices of the seven co-occurring plant species on the two islands, we visualized their plant–pollinator networks. Quantitative network metrics are more robust to sampling bias than qualitative networks, which are preferred for comparisons between communities. To examine the structure of the plant–pollinator networks, we calculated three network-level indices that are not sensitive to scale, with most of them being frequency-based and considering the frequency of interactions. Network connectance was calculated using the index of weighted connectance, which is the proportion of realized interactions in the network weighted by the interaction strength of each species. Network nestedness was calculated using the weighted nestedness metric based on overlap and decreasing fill (WNODF), which is a measure of the degree of nestedness for quantitative data, and it describes the extent to which species with fewer interactions are preferentially associated with a subset of species that interact with the most connected ones. Network specialization was calculated using the index H2′, which measures the level of interaction complementarity within the network. All network-level metrics were calculated using the “networklevel” function in the bipartite package. To ensure that sample size differences did not affect results, weighted connectance, WNODF, and H2' were bootstrapped 1,000 times using the bootstrapnet package. Overlapping niches are vital to maintain the complementarity in specialization of plant–pollinator interactions and provide pollinators with flexibility when foraging (Blüthgen & Klein, 2011). Given the importance of overlapping niches to plant–pollinator networks in redundancy of functional roles, we calculated the niche overlap of plants and pollinators that measures the extent to which plants share pollinators and vice versa using the “networklevel” function in the bipartite package (Dormann et al., 2021). Values of niche overlap vary from 0 (shared no interaction partners) to 1 (shared all interaction partners). We repeated the above analyses of plant–pollinator networks consisting of seven co-occurring plant species observed on both islands. To explore whether the role of specific pollinator functional groups differs in the plant–pollinator network structure, we repeated the above analyses for four subnetworks, including (1) Apidae, (2) non-Apidae Hymenoptera, (3) Diptera, and (4) Lepidoptera. Given the global importance of Apidae in plant–pollinator interactions and to further clarify the mechanisms within the diverse order of Hymenoptera, only the Apidae subnetwork was included.