Introduction
Specialization is a core dimension of the relationships between flowering plants and floral visitors (Armbruster Reference Armbruster2016; Waser and Ollerton Reference Waser and Ollerton2006). While ecological specialization has been defined in many ways in the literature (Devictor et al. Reference Devictor, Clavel, Julliard, Lavergne, Mouillot, Thuiller, Venail, Villéger and Mouquet2010), in the context of anthecology (the study of flowers, floral visitors, and their relationships), it broadly refers to how restrictive an organism is in terms of the partners it interacts with (Armbruster Reference Armbruster2016; Baker Reference Baker, Jones and Little1983; Robertson Reference Robertson1895). The degree to which either participant in a plant-pollinator mutualism is specialized has implications for how reliant those species are on each other, whether that be a plant’s dependency on a visitor as a pollinator, or a visitor’s dependency on a flower as a source of nutritional resources. As such, specialization is at the centre of ongoing research on population dynamics, competition, community assembly, response to disturbance, co-evolutionary trajectories, and speciation in both flowering plants and anthophilous animals (Armbruster Reference Armbruster2014, Reference Armbruster2016; Benadi et al. Reference Benadi, Blüthgen, Hovestadt and Poethke2013; Biesmeijer et al. Reference Biesmeijer, Roberts, Reemer, Ohlemüller, Edwards, Peeters, Schaffers, Potts, Kleukers, Thomas, Settele and Kunin2006; Lomáscolo et al. Reference Lomáscolo, Giannini, Chacoff, Castro-Urgal and Vázquez2019; Valdovinos and Marsland Reference Valdovinos and Marsland2021). Given the central role that specialization plays in our understanding of the ecology and evolution of flowers and their visitors, it is important that we be able to measure it accurately and consistently.
In recent decades, researchers have increasingly used bipartite network analyses to assess specialization in communities of flowering plants and anthophilous animals (Knight et al. Reference Knight, Ashman, Bennett, Burns, Passonneau and Steets2018). Anthecological systems are often highly diverse, and the relationships within them are complex. Analyzing these systems as ecological networks allows for the roles and positions of multiple taxa to be evaluated simultaneously, as well as enabling the investigation of emergent, system-level properties that might not be apparent from the study of individual taxa or smaller system sub-components (Bascompte Reference Bascompte2009; Vizentin-Bugoni et al. Reference Vizentin-Bugoni, Maruyama, De Souza, Ollerton, Rech, Sazima, Dáttilo and Rico-Gray2018). Methods for quantifying specialization in bipartite networks facilitate its evaluation at multiple levels of organization, including the whole network, either ‘party’ (i.e., either all the plants or all the visitors in the network), the guild/functional group (e.g. ‘bee’, ‘hummingbird’), and the individual species (Blüthgen et al. Reference Blüthgen, Menzel and Blüthgen2006; Fründ et al. Reference Fründ, McCann and Williams2016). These network-derived measures of specialization are integral to current research on a broad range of topics and provide the foundation for theoretical relationships between specialization and the dynamics of plant and pollinator communities. For example, network analyses have been used to show changes in community-level specialization in response to environmental drivers such as changing climate or land use (Chesshire et al. Reference Chesshire, McCabe and Cobb2021; Kaiser-Bunbury et al. Reference Kaiser-Bunbury, Mougal, Whittington, Valentin, Gabriel, Olesen and Blüthgen2017; Shinohara et al. Reference Shinohara, Uchida and Yoshida2019), and in simulated networks, more specialized organisms and communities are less resilient to disturbance and experience higher extinction rates (Clavel et al. Reference Clavel, Julliard and Devictor2011; Weiner et al. Reference Weiner, Werner, Linsenmair and Blüthgen2014).
However, because the bipartite networks used in anthecological studies are typically built from observations of visits to flowers and do not include information on what resources visitors are acquiring, estimates of specialization from these networks may not accurately depict visitors’ dependencies on floral hosts. In bipartite floral interaction networks, floral visitors and flowering plants are represented by nodes, which are connected by links that represent the presence and/or strength of a relationship between those nodes. In most published plant-pollinator network studies, the strength (or ‘weight’, in network parlance) of links is determined by rates of observed visitation and does not incorporate information on what resources visitors are acquiring. This is potentially problematic because floral visitors frequently require multiple floral resources, such as nectar, pollen, or floral oils, and may specialize to different degrees on each of the floral resources they use. For example, many bee species have specialized pollen diets, meaning they acquire pollen to provision their young from flowers of a limited set of plant taxa (typically referred to as ‘oligolectic’ bees, Danforth et al., Reference Danforth, Minckley, Neff and Fawcett2019; Waser and Ollerton, Reference Waser and Ollerton2006). However, adults of these oligolectic bees often visit flowers of additional species to search for the nectar they need to fuel their foraging activity (Cane and Sipes Reference Cane and Sipes2006; Robertson Reference Robertson1925). In a network built from observations of visitation, oligolectic bees can appear to be generalized, while in reality their persistence hinges on access to a narrow subset of the flowers they visit (e.g. Gibson et al. Reference Gibson, Onuferko, Myers and Colla2024). Similar variation in resource-specific specialization may occur in other taxa that use multiple floral resources, such as the many flower-visiting beetle, fly, and bug species (Haslett Reference Haslett1989; Wäckers et al. Reference Wäckers, Romeis and Van Rijn2007; Wedmann et al. Reference Wedmann, Hörnschemeyer, Engel, Zetter and Grímsson2021), although knowledge on floral resource specialization in these groups is relatively sparse.
While most anthecologists will be aware of examples where observed patterns of floral visitation are a poor proxy for the ecological specialization of a flower-visiting taxon, investigations of how this impacts the inferences made from floral visitation networks have been limited. Given the widespread use of network-derived measures of specialization, it is important to evaluate whether estimates of visitors’ specialization from visitation networks are representative of their actual specialization on the resources they are foraging for, and how this propagates to community-level measures of specialization. One approach to answering this question is to compare visitation networks to pollen-use networks inferred from the pollen found on visitors’ bodies (Tourbez et al. Reference Tourbez, Gómez-Martínez, González-Estévez and Lázaro2024), or in bee nests (Dorado et al. Reference Dorado, Vázquez, Stevani and Chacoff2011; Gresty et al. Reference Gresty, Clare, Devey, Cowan, Csiba, Malakasi, Lewis and Willis2018). However, direct comparisons of visitation networks to pollen-identification networks are complicated by difficulties in ensuring that both networks encapsulate the same spatial scale (Bosch et al. Reference Bosch, Martín González, Rodrigo and Navarro2009), are sampled to a similar level of completeness (De Manincor et al. Reference De Manincor, Hautekèete, Mazoyer, Moreau, Piquot, Schatz, Schmitt, Zélazny and Massol2020), the fact that non-bee pollinators do not generally carry pollen destined for consumption externally, and the contamination of body pollen from sources that are not associated with pollen foraging (Vaknin et al. Reference Vaknin, Gan-Mor, Bechar, Ronen, Eisikowitch, Dafni, Hesse and Pacini2000). Furthermore, pollen-identification networks offer little insight into nectar specialization, and many floral visitors discriminate between nectar sources (Cronk and Ojeda Reference Cronk and Ojeda2008; Heinrich Reference Heinrich2004). An analogous question from the plants’ point of view, of how well visitation approximates specialization on specific pollinators, has been investigated several times (Ballantyne et al. Reference Ballantyne, Baldock and Willmer2015, Reference Ballantyne, Baldock, Rendell and Willmer2017; Cirtwill et al. Reference Cirtwill, Wirta, Kaartinen, Ballantyne, Stone, Cunnold, Tiusanen and Roslin2024; De Santiago Hernández et al. Reference De Santiago-Hernández, Martén-Rodríguez, Lopezaraiza-Mikel, Oyama, González-Rodríguez and Quesada2019; Jędrzejewska-Szmek and Zych Reference Jędrzejewska-Szmek and Zych2013). These phytocentric studies typically find that visitation networks have more connections than networks of pollen deposition, and that plants in actuality rely on a subset of visiting species to achieve successful pollination. Thus, plants are more specialized than is implied by visitation data alone.
In this study, we use observations of floral interactions from the canopy of a tropical rainforest to compare the structure of a network built on visitation to that of networks built using only records of nectar or pollen foraging, respectively. Our study and data set are novel in several ways. Previously, only a handful of studies globally (across both temperate and tropical zones) have applied a network approach to study floral interactions in the forest canopy, and/or evaluated specialization of canopy pollinators (Swart et al. Reference Swart, Geerts, Pryke and Coetzee2024; Wardhaugh et al. Reference Wardhaugh, Edwards and Stork2015). Furthermore, the Asian tropics are generally under-represented in the plant-pollinator network literature (Vizentin-Bugoni et al. Reference Vizentin-Bugoni, Maruyama, De Souza, Ollerton, Rech, Sazima, Dáttilo and Rico-Gray2018) (but see Hass et al. Reference Hass, Liese, Heong, Settele, Tscharntke and Westphal2018; Simla et al. Reference Simla, Chaianunporn, Sankamethawee, Hughes and Sritongchuay2022; Sritongchuay et al. Reference Sritongchuay, Hughes and Bumrungsri2019a, Reference Sritongchuay, Hughes, Memmott and Bumrungsri2019b for recent ground-level networks from the region). Historically, most tropical canopy plant pollinator research has been conducted using the handful of fixed canopy-access structures scattered around the world (e.g. Roubik Reference Roubik1993; Roubik et al. Reference Roubik, Sakai, Gattesco, Lowman and Rinker2003, Reference Roubik, Sakai and Karim2005; Van Dulmen Reference Van Dulmen2001). By using rope-access techniques, we expand our research to a habitat type (tropical peat swamp forest) lacking such infrastructure, and to larger spatial scales than are accessible from fixed structures.
Specifically, we use our floral interaction data set to ask the following questions:
-
1. How different are estimates of flower visitor specialization when they are based on a visitation network versus a nectar-foraging or pollen-foraging network?
-
2. How different are estimates of specialization for the visitor community as a whole when they are based on a visitation network versus a nectar-foraging or pollen-foraging network?
Materials and methods
Study system and species
This study uses floral interaction data collected in the canopy and sub-canopy of a tropical peat swamp forest in Central Kalimantan, a province in Indonesian Borneo. Observations of floral interactions were made on locally occurring species of tree and woody shrub in the genus Syzygium (Myrtaceae, P. Browne ex Gaertn.). Syzygium is one of the world’s most diverse genera of woody plant (Ahmad et al. Reference Ahmad, Baider, Bernardini, Biffin, Brambach, Burslem, Byng, Christenhusz, Florens, Lucas, Ray, Ray, Smets, Snow, Strijk and Wilson2016; Low et al. Reference Low, Rajaraman, Tomlin, Ahmad, Ardi, Armstrong, Athen, Berhaman, Bone, Cheek, Cho, Choo, Cowie, Crayn, Fleck, Ford, Forster, Girmansyah, Goyder, Gray, Heatubun, Ibrahim, Ibrahim, Jayasinghe, Kalat, Kathriarachchi, Kintamani, Koh, Lai, Lee, Leong, Lim, Lum, Mahyuni, McDonald, Metali, Mustaqim, Naiki, Ngo, Niissalo, Ranasinghe, Repin, Rustiami, Simbiak, Sukri, Sunarti, Trethowan, Trias-Blasi, Vasconcelos, Wanma, Widodo, Wijesundara, Worboys, Yap, Yong, Khew, Salojärvi, Michael, Middleton, Burslem, Lindqvist, Lucas and Albert2022), and it reaches peak diversity in the tropical rainforests of southeast Asia, where it is the most abundant and diverse tree genus in many types of forest (Cannon and Lerdau Reference Cannon and Lerdau2015; Slik et al. Reference Slik, Poulsen, Ashton, Cannon, Eichhorn, Kartawinata, Lanniari, Nagamasu, Nakagawa, Van Nieuwstadt, Payne, Purwaningsih, Saridan, Sidiyasa, Verburg, Webb and Wilkie2003). The open and accessible brush-like flowers are used as a source of pollen and nectar by a remarkably broad range of anthophilous taxa, including birds, bees, beetles, and butterflies (Boulter et al. Reference Boulter, Kitching, Howlett and Goodall2005; Kuriakose et al. Reference Kuriakose, Sinu and Shivanna2018; Lughadha and Proenca Reference Lughadha and Proenca1996). In Bornean peat swamp forests, the flowering of Syzygium species is not seasonally restricted, instead providing irregular resource pulses year-round (Harrison et al. Reference Harrison, Zweifel, Husson, Cheyne, D’Arcy, Harsanto, Morrogh-Bernard, Purwanto, Rahmatd, Vogel, Wich and Noordwijk2016), including during seasons where alternative floral resources are lacking (as high as 38% of flowering trees during months with low overall flowering, unpubl. data). Thus, Syzygium species represent a key resource for the anthophilous communities within peat swamp forests.
We decided to restrict data collection to this single genus for several reasons. First, the deep hypanthium cup and filamentous stamens of Syzygium flowers facilitate the classification of resource use by floral visitors, enabling differentiation between flower visitors foraging for pollen versus nectar (Figure 1). Second, different species flower throughout the year, their flowers are easy to spot, and they attract high rates of visitation, enabling continuous data collection over the study period. Third, the wood qualities and branching architecture of Syzygium species facilitate safe access to canopy inflorescences, unlike many tropical peat swamp forest tree species. Studying a network restricted to interactions within a single plant genus may result in different estimates of visitor specialization than studying interactions across a whole community of plants, or even across a less closely related subset of the community, as pollinators often specialize on specific plant lineages while demonstrating generalized foraging patterns within those lineages (Robertson Reference Robertson1925; Waser and Ollerton Reference Waser and Ollerton2006). However, the main objective of this work is not to estimate specialization for specific taxa or to compare specialization across different communities, but rather to explore how different approaches to building networks can influence our understanding of specialization in a single community.
Figure 1. Examples of pollen and nectar foraging on Syzygium flowers. Clockwise from top-left: a. Hylaeus penangensis Cockerell and Ceratina cf. nigrolateralis foraging for pollen and nectar, respectively; b. nose fly (Rhiniidae sp.) consuming pollen; c. Olive-backed Sunbird (Cinnyris jugularis L.) nectaring; d. Scarlet-backed Flowerpecker (Dicaeum cruentatum L.) nectaring on Syzygium oligmyrum Diels; and e. flower chaffer (Taeniodera sp.) consuming Syzygium pollen.
Data collection was conducted in the vicinity of the Tuanan Research Center, which is located within the recently listed Mawas Protected Area, one of the largest remnant patches of peat swamp forest in Central Kalimantan. Study trees were distributed across a mixture of primary peat swamp forest, recovering (post-fire) forest edge, and recently deforested/burnt habitat. Each individual study tree was assigned to a species by experts in the genus (see Acknowledgements), although determining the correct scientific binomial was not always possible given the unresolved state of taxonomy for the genus in the region (Ahmad et al. Reference Ahmad, Baider, Bernardini, Biffin, Brambach, Burslem, Byng, Christenhusz, Florens, Lucas, Ray, Ray, Smets, Snow, Strijk and Wilson2016).
Few studies of floral visitors have been conducted in Bornean peat swamp forests, and as such, their identities and taxonomies remain largely unknown. Vertebrate visitors, including nectarivorous birds in several families and nectarivorous bats in the Pteropodidae, are important pollinators in many Bornean forests (Kato Reference Kato1996; Roubik et al. Reference Roubik, Sakai and Karim2005; Sakai Reference Sakai2000). They tend to be well described and can consistently be identified to the species level (Phillipps and Phillipps Reference Phillipps and Phillipps2014, Reference Phillipps and Phillipps2016). Conversely, invertebrate floral visitors at Tuanan are likely to belong to poorly understood groups with incomplete taxonomies, for which comprehensive identification resources don’t exist. Indeed, a few recent invertebrate surveys at Tuanan have encountered species new to science, as well as new records for the region (Dow and Silvius Reference Dow and Silvius2014; Widowati et al. Reference Widowati, Maulana and Atmoko2023, Issa Bettencourt pers. comm.)
Data collection
Between October 2022 and July 2023, AA observed visitation and resource-foraging on flowers of 15 co-occurring Syzygium species at Tuanan (Table 1). Walking surveys for blooming trees were conducted weekly, and once discovered, trees were selected for inclusion based on ease-of-access, safety for climbing, and the amount of data already collected for that species. Ultimately, 37 individual flowering trees located within a 1.6 km radius of the station were included in the study. Observations were carried out from shortly after sunrise to shortly after sunset, as long as it was not raining. Typically, 2–3 trees were visited during a field day with good weather. Visitors to Syzygium flowers were observed in 30-minute periods, during which all visitors to a clump of 1–3 inflorescences (5–200 flowers) were recorded. Visitors were assigned to consistently identifiable and mutually exclusive visual morphogroups (Table 2). For each visitor morphogroup, we recorded the number of individuals that visited flowers (defined as any visitor that touched the perianth or floral reproductive parts), and among those, how many foraged for pollen and/or nectar. As such, the pollen- and nectar-foraging data are subsets of the visitation data. A total of 152 such observation periods were conducted (76 hours). Observations were conducted at a maximum distance of 3 m from the inflorescences, using insect binoculars (Pentax Papilio II 6.5×21). Most observations were conducted in the canopy, which was accessed using ropes. On the forest edge, where flowers extend to near ground level, and on some shrubby Syzygium species, observations were conducted from the ground. Immediately after visual observations, voucher specimens of invertebrate visitors were collected from the same part of the tree for 15 minutes (total netting time, excluding processing time), to provide higher taxonomic resolution information on visual morphogroup composition (Table S2).
Table 1. Syzygium trees in included in this study. Syzygium species were identified by AA, Dr. Yee Wen Low (Singapore Botanic Gardens), Dr. Peter Ashton (Arnold Arboretum, Harvard University, emeritus), and Bina Swasta Sitepu (Wanariset Harbarium) from pressed specimens. Pressed specimens for each tree included in the study were deposited at the Wanariset Herbarium in Samboja, Indonesia
Table 2. Visitor morphogroups used in this study. All groups are mutually exclusive (e.g. ‘other flies’ does not include members of the family Culicidae). Tax. res. indicates the smallest taxonomic level that encompasses all group members. Information on the composition of morphogroups according to collected voucher specimens can be found in supplementary Tables S2–3
In the first days of canopy observations, it became apparent that vertebrate visitors were affected by the observer’s presence in the canopy. Thereafter, all observations of vertebrate visitors were conducted from the ground or lower canopy using 10×42 birding binoculars, either prior to climbing into the canopy to observe the insects or after canopy work was completed. In these vertebrate observation rounds (100 in total, or 50 hours), which were separate from and in addition to the invertebrate observation rounds, all vertebrates that drank nectar from flowers on the visible part of the tree were recorded. No vertebrates were observed to consume pollen or purposefully interact with stamens and anthers.
To assess the potential importance of nocturnal flower visitors such as bats and moths, AA conducted 13 hours of crepuscular/nocturnal observation in 2022–2023. Nocturnal invertebrate and vertebrate observation rounds were conducted in the same way as their diurnal counterparts, including canopy observations, but a high-power red-light headlamp was used to illuminate inflorescences. Bats and most nocturnal invertebrates (including moths) have been shown not to respond to red light sources (Brehm et al. Reference Brehm, Niermann, Jaimes Nino, Enseling, Jüstel, Axmacher, Warrant and Fiedler2021; Briscoe and Chittka Reference Briscoe and Chittka2001; Spoelstra et al. Reference Spoelstra, Van Grunsven, Ramakers, Ferguson, Raap, Donners, Veenendaal and Visser2017), minimizing behavioural interference. No nocturnal vertebrate pollinators were observed, and mean rates of invertebrate visitation were very low (<4 per hour, compared with ∼23 per hour for diurnal sampling). Nonetheless, to ascertain the absence of nocturnal vertebrate pollinators, field workers conducted an additional 36 hours of nocturnal observations from July to December 2024, using red-light headlamps and night-vision cameras, during which no nocturnal vertebrate visitors were recorded. The 2024 observations were limited to the subcanopy and or canopies on the forest edge with a clear line-of-sight from the ground.
Analytical approach
To investigate how estimates of specialization from resource-indiscriminate visitation data compare to estimates of specialization based exclusively on nectar or pollen foraging, we built three separate weighted (i.e., including information about interaction frequency) bipartite networks and calculated the same set of specialization metrics for each. In the first network, weights of interactions were determined by the number of times a visitor morphogroup (Table 2) was observed visiting a Syzygium species (Table 1, see Tables S1–S3 for detailed information on Syzygium species and visitor morphogroup compositions). In the two resource-specific networks, only visits that included the observed use of either nectar or pollen were included (‘nectar-foraging’ and ‘pollen-foraging’ networks, respectively).
To compare visitor specialization across networks, we calculated d’, a commonly used specialization metric, for each visitor morphogroup in each network (Blüthgen et al. Reference Blüthgen, Menzel and Blüthgen2006). For visitor nodes (which typically represent a visitor taxon, such as a species or genus), d’ is a measure of the Shannon diversity of the ‘community’ of plant species it interacts with, rescaled between 0 (least specialized) and 1 (most specialized). A visitor node’s specialization is assessed relative to the ‘availability’ of the plant species in the network, which is measured as the frequency with which each plant is visited by all the visitors in the network. A d’ of 0 signifies that a flower visitor interacts with plant species exactly proportionally to how often they interact with all the visitors in the network (i.e. proportionally to their weighted degrees). A high d’ indicates that a visitor species interacts with plant species non-randomly, and typically corresponds to the observed use of just one interaction partner. Since plants’ weighted degrees are calculated independently in each of the three networks we consider here (all visitation, nectar foraging, and pollen foraging), a visitor species’ resource-specific foraging may be more proportional to availability for one resource, and therefore score as less specialized (lower d’), even if it visits fewer Syzygium species for that resource than it does for the other. This differs from classical measures of specialization based on raw counts of the number of interaction partners (Armbruster Reference Armbruster2016; Waser and Ollerton Reference Waser and Ollerton2006), where a visitor would not be able to measure as less specialized in a network where it had fewer interaction partners. The classical, count-based metrics are highly sensitive to sample size (i.e., the number of plant-pollinator interactions observed in the network), which has led to their decline in use in favour of d’ and related measures (Dormann Reference Dormann2011; Fründ et al. Reference Fründ, McCann and Williams2016).
After computing d’ for each visitor morphogroup in each network, we calculated the difference in visitor morphogroup specialization values between the nectar-foraging network and the visitation network (d’nectar – d’visits = Δd’nectar) and between the pollen-foraging network and the visitation network (d’pollen – d’visits = Δd’pollen) (Figure 2). We also tested whether morphogroups were significantly more or less specialized, on average, in the resource-specific networks than the visitation network using paired t-tests on the d’ values of morphogroups that were present in both the visitation network and each resource-specific network. Additionally, because raw values of d’ are difficult to interpret ecologically, we graphically demonstrate differences in how each network ranks visitor morphogroups in terms of specialization in Figure 3.
Figure 2. Boxplots of the difference in visitor morphogroup specialization between resource-specific networks and the visitation network. The y-axis values are the differences in d’ value for the same morphogroups between the visitation and the nectar foraging networks (d’nectar – d’visits) and the visitation and the pollen foraging networks (d’pollen – d’visits), respectively. Each point represents a single visitor morphogroup. Nectar specialization was significantly higher than visitation specialization on average (paired t-test p value = 0.005), while pollen specialization often differed in magnitude, but was not significantly higher or lower on average (paired t-test p value = 0.648).
Figure 3. Visitor morphogroups ranked in decreasing order of specialization in each network. Fill colour represents relative specialization in the visitation network (orange = most specialized, blue = least specialized). Lines connect the same visitor morphogroups between networks. Only the 15 morphogroups that foraged on both nectar and pollen are included.
To assess how differences in estimates of visitor morphogroup specialization scale up to affect estimates of specialization at higher levels of biological organization, we calculated the weighted mean d’ for the entire flower visitor community in each network (<d’visitors>, following Blüthgen et al. Reference Blüthgen, Menzel and Blüthgen2006). Because this community-level metric can be sensitive to differences in network size (i.e., differences in the number of included taxa and their interaction totals) (Fründ et al. Reference Fründ, McCann and Williams2016), which varied across our three networks, we generated null expectations of specialization for the visitation network at different sizes by randomly removing interactions and recalculating <d’visitors>. We then compared the observed <d’visitors> for each resource-specific network to the distribution of rarefied null <d’visitors> values (Figure 4). The whole-network equivalent of <d’visitors>, which includes both plants and visitors and is known as H2’, is more frequently encountered in the literature. However, we opt to report <d’visitors> here because it is more directly related to our question of visitor specialization. Evaluating H2’ instead of <d’visitors> did not qualitatively change the results (Figure S1).
Figure 4. <d’visitors> (visitor community weighted mean d’) plotted against network size (total number of observed interactions) for the visitation, nectar-foraging, and pollen-foraging networks. Error bars represent the inner 95% of <d’visitors> values for 99 rarefactions of the visitation network repeated at 20-interaction intervals.
All analyses were conducted in R (R Core Team 2023), and the following packages were used to conduct analyses and generate figures: dplyr, ggplot2, bipartite, igraph, tidyr, forcats, ggpubr, rstatix (Csárdi et al. Reference Csárdi, Nepusz, Müller, Horvát, Traag, Zanini and Noom2024; Dormann et al. Reference Dormann, Gruber and Fründ2008; Kassambara Reference Kassambara2023a, Reference Kassambara2023b; Wickham Reference Wickham2016, Reference Wickham2023; Wickham et al. Reference Wickham, François, Henry, Müller and Vaughan2023a, Reference Wickham, Vaughan and Girlich2023b).
Results
We observed a total of 1977 visits by individuals from 28 visitor morphogroups to 15 Syzygium species. Individuals from all 28 visitor morphogroups foraged on nectar, and all 15 Syzygium species served as a source of nectar for at least one visitor morphogroup. A total of 1112 nectar foraging visits were observed. The pollen-foraging network was smaller, with 614 pollen foraging visits by 15 visitor morphogroups to 11 Syzygium species. Identification of vertebrate visitors in the field and of voucher specimens of invertebrates in the lab resulted in 297 unique visitor species and morphospecies (Tables S2–3). Invertebrate visitors were predominant, with 766 visits by bees (Hymenoptera: Anthophila), 337 by beetles (Coleoptera), 200 by ants (Hymenoptera: Formicidae), 157 by wasps (Hymenoptera), 155 by flies (Diptera), and a handful of visits by a diversity of other insect orders (e.g. Lepidoptera, Blattodea, Hemiptera). Floral visits by invertebrates that were exclusively ambush predators and did not consume nectar or pollen were excluded from the networks (Mantodea [4 visits] and Salticidae [1]), as were visits by invertebrate orders with fewer than 5 total recorded visits (Orthoptera [2] and Dermaptera [1]). Vertebrate observation rounds recorded 260 visits by birds from several families and one species of squirrel (Table S3).
Visitors in the nectar-foraging network were more specialized (paired t-test p = 0.005), and visitors in the pollen-foraging network similarly specialized (paired t-test p = 0.648), as compared with the network based on all flower visitation (Figure 2). For the 28 visitor morphogroups included in the nectar-foraging network, mean Δd’nectar was 0.04 (range: –0.07 to 0.17), indicating that they were, on average, more specialized for nectar foraging than for all flower visitation. Overall, 71% of morphogroups (20/28) were more specialized in the nectar-foraging network than they were in the all visitation network. For the 15 visitor morphogroups included in the pollen-foraging network, mean Δd’pollen was –0.02 (range: –0.24 to 0.2), indicating that species were, on average, less specialized for pollen foraging than for all flower visitation. Overall, 40% (6/15) of morphogroups were less specialized in the pollen-foraging network than they were in the all visitation network.
One way to assess how much accounting for the type of floral resource a visitor forages on during visits affects estimates of its specialization is to ask whether the rank order of groups, from most to least specialized, changes across the interaction networks. For the 15 visitor morphogroups observed foraging on both pollen and nectar, we compared rankings based on the all visitation network with those based on resource-specific networks. Most morphogroups changed rank across these networks, and in many cases, the shifts were substantial. On average, the same morphogroup’s specialization rank differed by 3.3 positions between any two networks, with differences as large as 12 ranks (Figure 3). These shifts indicate that a visitor considered relatively generalized when all visits are treated equally might appear much more specialized when only pollen or nectar foraging is considered.
The visitor community as a whole was more specialized in the nectar-foraging network than it was in all visitation network (<d’poll> = 0.287 and 0.253 for the observed nectar-foraging and all visitation networks, respectively). Comparison to rarefaction null visitation networks of the same size as the nectar-foraging network found the difference in specialization to be highly significant (p ≤ 0.001) (Figure 4). The visitor community also scored as more specialized in the pollen-foraging network than in the all visitation network (<d’poll> = 0.277 in the pollen-foraging network). However, that score was well within the range of rarefied null visitation networks of the same size as the pollen-foraging network (p = 0.90), indicating the difference in specialization was non-significant.
Discussion
The extent to which a flower-feeding animal is specialized — that is, how restrictive it is in terms of the diversity of flowering plant species it acquires resources from — is an important dimension of its ecology and natural history. Yet the plant-pollinator networks used to evaluate specialization are typically based upon observations of flower visits, and do not distinguish whether the flower visitor is foraging for pollen or nectar. In reality, visitors may be more selective about where they acquire one or the other resource, creating an opportunity for resource-indiscriminate visitation data to misrepresent visitors’ dependencies on the plants they visit. Research on the analogous potential for visitation networks to misrepresent plants’ dependencies on visitors as pollinators has shown that plants are frequently more specialized than is implied by visitation networks (Ballantyne et al. Reference Ballantyne, Baldock and Willmer2015, Reference Ballantyne, Baldock, Rendell and Willmer2017; De Santiago-Hernández et al. Reference De Santiago-Hernández, Martén-Rodríguez, Lopezaraiza-Mikel, Oyama, González-Rodríguez and Quesada2019; King et al. Reference King, Ballantyne and Willmer2013). Surprisingly, given the attention devoted to the phytocentric perspective, analogous investigations from the visitors’ point of view have been limited.
We accessed the canopy of a tropical rainforest to collect a floral interaction dataset that differentiates pollen and nectar foraging, and compared how a traditional visitation-based network differed from networks built on pollen- or nectar-foraging in their estimates of floral visitor specialization. We found that a floral interaction network built on resource-agnostic visitation data resulted in markedly different estimates of visitor specialization than did networks that reflect resource use (Figures 2 and 3). Likewise, specialization as measured for the entire flower-visitor community differed between the visitation-based network and the pollen- and nectar-foraging networks (Figure 4). Previously, significant differences between visitation-based and resource-specific specialization have been found when comparing visitation networks to networks inferred from the scopal pollen loads of bees (Tourbez et al. Reference Tourbez, Gómez-Martínez, González-Estévez and Lázaro2024). However, it has been unclear to what degree observed differences in specialization between visitation networks and pollen-identification networks is attributable to bees foraging for pollen unevenly on the plants they visit, as opposed to the tendency for pollen identification to provide more complete sampling of interactions (De Manincor et al. Reference De Manincor, Hautekèete, Mazoyer, Moreau, Piquot, Schatz, Schmitt, Zélazny and Massol2020), especially for rare visitor species (Dorado et al. Reference Dorado, Vázquez, Stevani and Chacoff2011). By using a system where nectar and pollen foraging can be differentiated during visits, we minimize the influence of sampling at different scales and levels of completeness, expand our analysis beyond the realm of bees, and compare nectar-specific foraging networks to pollen-foraging and visitation-derived networks for the first time.
As well as differences in the estimated values of specialization between networks, there were pronounced differences among our networks in terms of how flower visitors were ranked for specialization. The most specialized morphogroups in either of the resource-foraging networks were often amongst the more generalized morphogroups in the visitation network, and vice versa. This is well exemplified by the bee genus Nomia Latreille. Of the 15 visitor morphogroups that foraged on Syzygium for both pollen and nectar, Nomia were the least specialized nectar foragers, firmly in the middle of the pack in the visitation network (9th most specialized), and the 3rd most specialized pollen foragers (Figure 3). These findings highlight the potential for visitation networks to misidentify the most specialized taxa in a community of floral visitors.
Most visitors were more specialized in the nectar-foraging network than they were in the visitation network. The tendency for visitors to be more specialized nectar-foragers carried through to affect estimates of specialization for the whole visitor community, with the result that nectar-foraging was more specialized than either pollen-foraging or visitation at the network level. While the sign of the difference in specialization between the pollen-foraging and visitation networks was less predictable, the magnitudes of those differences were often large, with visitor morphogroups differing by as much as 24% of the maximum range of the d’ metric between the two. Though our findings appear to contradict the established theory that pollen-foraging is, on average, more specialized than nectar foraging (at least for bees) (Cane and Sipes Reference Cane and Sipes2006; Danforth et al. Reference Danforth, Minckley, Neff and Fawcett2019; Strickler Reference Strickler1979), it is perhaps not that surprising for a network which includes members of only one plant genus. From the point of view of pollen-foragers, Syzygium flowers may be largely interchangeable, with similar floral morphology to deal with when collecting pollen, and potentially similar pollen nutrient composition (insofar as pollen composition may be a phylogenetically conserved trait [Ruedenauer et al. Reference Ruedenauer, Spaethe, Van Der Kooi and Leonhardt2019], though as far as we are aware that has not been studied in the Myrtaceae). Conversely, we observed substantial variation in per-flower nectar volume and nectar sugar concentrations amongst the Syzygium species in this study (Figure S2). Resulting differences in the economics of foraging on different Syzygium flowers may explain the increased specialization in nectar foraging (Heinrich Reference Heinrich2004; Roubik Reference Roubik, Edwards, Booth and Choy1996). Further studies that consider a wider taxonomic range of plant species are needed to establish the generality of our findings.
The inability of the undifferentiated visitation data to accurately represent the resource-specialization of floral visitors questions the findings of a large body of network-based literature on specialization in plant and pollinator communities. Interestingly, many early studies on specialization in anthophilous animals did distinguish between floral resources, including foundational research by 19th and early 20th-century anthecologists (Robertson Reference Robertson1925; Waser Reference Waser2006). For instance, early research on temperate bees showed that they tended to be more selective about sources of larval provisions (typically pollen) than they were about sources of nectar. As a result, bees are defined as specialists or generalists based on the diversity of larval provision sources alone (Cane and Sipes Reference Cane and Sipes2006; Danforth et al. Reference Danforth, Minckley, Neff and Fawcett2019; Strickler Reference Strickler1979). However, with the advent of network approaches easier-to-collect visitation data have become the norm, and these data are now frequently used to compare specialization both within the same network and across different visitor communities (Petanidou and Potts Reference Petanidou, Potts, Waser and Ollerton2006, provides an example of both). The growing availability of published visitation networks has also stimulated meta-analysis of network-derived specialization metrics (Knight et al. Reference Knight, Ashman, Bennett, Burns, Passonneau and Steets2018; López-Vázquez et al. Reference López-Vázquez, Lara, Corcuera, Castillo-Guevara and Cuautle2024; Vizentin-Bugoni et al. Reference Vizentin-Bugoni, Maruyama, De Souza, Ollerton, Rech, Sazima, Dáttilo and Rico-Gray2018). Because, as we have shown, the links between plants and animals in a visitation network may not be representative of their true interdependencies, comparisons across systems that provide different resources, or between taxa with different floral resource requirements, may be affected. For example, a global comparison of specialization of insect pollinator taxa found that in warm climates bee species had a broader set of floral interaction partners, on average, than other insect pollinators (Saunders et al. Reference Saunders, Kendall, Lanuza, Hall, Rader and Stavert2023). However, because bee species on average visit flowers for more different resources (including pollen and nectar, but in some cases also floral oils or volatiles) than species from other insect groups included in the analysis (such as wasps or butterflies, which rarely consume pollen), there is more opportunity for them to appear generalized while in fact specializing on a more restricted set of plants for at least one floral resource.
The literature on geographic gradients in plant and pollinator specialization also relies heavily on visitation networks, despite clear differences across latitudes and regions in the diversity of floral resource types provided to floral visitors. As noted by Armbruster (Reference Armbruster, Waser and Ollerton2006), pollen and nectar rewards are ubiquitous across latitudes, whereas floral oils and brood-site provision are restricted to sub-arctic latitudes, and the provision of fragrant volatiles or resins as rewards is restricted to the tropics and subtropics. Consequently, opportunities for specialized mutualisms to be masked by resource-agnostic visitation data may increase with decreasing latitude. There are also trends in the identities and resource requirements of major pollinating taxa (Armbruster Reference Armbruster, Waser and Ollerton2006), which may affect biases in visitation networks—for example, primarily nectivorous bats and birds are absent from higher latitudes, and there are often strong altitudinal gradients in the functional diversity of pollinator taxa (Dellinger et al. Reference Dellinger, Hamilton, Wessinger and Smith2023). Thus, conclusions that pollinator communities at lower latitudes (Saunders et al. Reference Saunders, Kendall, Lanuza, Hall, Rader and Stavert2023; Schleuning et al. Reference Schleuning, Fründ, Klein, Abrahamczyk, Alarcón, Albrecht, Andersson, Bazarian, Böhning-Gaese, Bommarco, Dalsgaard, Dehling, Gotlieb, Hagen, Hickler, Holzschuh, Kaiser-Bunbury, Kreft, Morris, Sandel, Sutherland, Svenning, Tscharntke, Watts, Weiner, Werner, Williams, Winqvist, Dormann and Blüthgen2012) and lower elevations (Olesen and Jordano Reference Olesen and Jordano2002) are more generalized should be revisited, taking into account correlated trends in diversity of resources provided by flowers and utilized by visitors.
By using data collected in the canopy of a Southeast Asian tropical forest, this study helps fill geographic, biome, and forest-stratal gaps in the plant-pollinator network literature (Vizentin-Bugoni et al. Reference Vizentin-Bugoni, Maruyama, De Souza, Ollerton, Rech, Sazima, Dáttilo and Rico-Gray2018; Zanata et al. Reference Zanata, Dalsgaard, Passos, Cotton, Roper, Maruyama, Fischer, Schleuning, Martín González, Vizentin-Bugoni, Franklin, Abrahamczyk, Alárcon, Araujo, Araújo, Azevedo-Junior, Baquero, Böhning-Gaese, Carstensen, Chupil, Coelho, Faria, Hořák, Ingversen, Janeček, Kohler, Lara, Las-Casas, Lopes, Machado, Machado, Machado, Maglianesi, Malucelli, Mohd-Azlan, Moura, Oliveira, Oliveira, Ornelas, Riegert, Rodrigues, Rosero-Lasprilla, Rui, Sazima, Schmid, Sedláček, Timmermann, Vollstädt, Wang, Watts, Rahbek and Varassin2017). We found only two previous studies that applied a network approach to studying floral interactions in the forest canopy, both of which also used Bluthgen’s family of network specialization metrics to evaluate the specialization of canopy pollinators. Wardhaugh et al. (Reference Wardhaugh, Edwards and Stork2015) used a canopy crane in tropical forests in northern Australia to study beetle assemblages and found similar degrees of specialization in anthophilous and folivorous beetles, contrary to theoretical expectations. Swart et al. (Reference Swart, Geerts, Pryke and Coetzee2024) used rope access to study network structure and specialization amongst visitors to four tree species with similar floral morphology in a temperate montane forest in South Africa, using observations of visits to estimate pollinator specialization (pollen from visitor bodies was also identified, but not used to estimate specialization of visitor taxa). While most pollinators visited all four tree species and scored relatively low (highly generalized) values of d’, they showed clear preferences for some trees over others, despite the generalized morphology of the flowers. The latter study, along with the work we present here, is representative of a growing movement amongst anthecologists to escape the confines of fixed canopy infrastructure by using elevated traps (Roubik Reference Roubik1993; Ulyshen et al. Reference Ulyshen, Soon and Hanula2010; Urban-Mead et al. Reference Urban-Mead, Muñiz, Gillung, Espinoza, Fordyce, Van Dyke, McArt and Danforth2021), canopy cameras (Droissart et al. Reference Droissart, Azandi, Onguene, Savignac, Smith and Deblauwe2021), or rope access (Reyes et al. Reference Reyes, Draper and Marques2021; Swart et al. Reference Swart, Geerts, Pryke and Coetzee2024), which has led to significant gains of knowledge on canopy pollination and pollinator ecology in recent years. Nonetheless, despite the increasing flexibility offered by these methods and the proliferation of anthecological research in recent decades (Knight et al. Reference Knight, Ashman, Bennett, Burns, Passonneau and Steets2018), as far as we are aware this study represents the first published plant-pollinator network study from the canopy of an Asian forest, the first study of floral interactions in a Bornean peat swamp forest, and one of only a handful of published plant-pollinator networks from tropical Asia, generally.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0266467425100175
Acknowledgements
First, thank you to Dr. Erin Vogel for encouraging and enabling this work to be done at the Tuanan Research Station. We thank Dr. John S. Ascher (National University Singapore), Matthew T. Hamer (University of Hong Kong), Dr. Jon Gelhaus (Drexel University), and members of the iNaturalist community for aid in identifying invertebrate specimens. Thank you to Dr. Yee Wen Low (Singapore Botanic Gardens), Bina Swasta Sitepu (Wanariset Herbarium), and Dr. Peter S. Ashton (Harvard University, emeritus) for aid in identifying Syzygium species. We greatly appreciate the support of M. Rizqi Arifuddin and everyone at the American Indonesian Exchange Foundation (AMINEF) for their support with logistics in Indonesia. Thank you to Pak Uteo for his assistance in the field, as well as the rest of the Tuanan research staff, and the Tuanan community for their support of research efforts at Tuanan Research Station. Thank you to Encilia, Pak Mesa, Pak Sih Kahono, and other BRIN scientists for facilitating our work at the Research Center for Biosystematics and Evolution (BRIN-PRBE). We gratefully acknowledge the Indonesian State Ministry for Research and Technology (RISTEK-BRIN), the Indonesian Institute of Science (BRIN), the Directorate General of Forest Protection and Nature Conservation (KSDAE Kementerian Kehutanan), Ministry of Internal Affairs, the Nature Conservation Agency Palangkaraya (BKSDA), the local government in Central Kalimantan (KPHL Kapuas-Kahayan and Dinas Kehutanan), the Borneo Orangutan Survival Foundation, and BOS-Mawas for their permission and support to the project. Thank you also to members of the Winfree lab at Rutgers University, who provided feedback on an early draft of the manuscript, and Dr. Neal Williams (UC Davis), who reviewed a later draft.
Financial support
A.A. was supported by a Fulbright U.S. Student Scholarship, which is funded by the US Department of State and AMINEF, for the duration of his time in the field. The Tuanan Research Station is supported by the National Science Foundation (BCS-1719825), Rutgers University, and the Borneo Orangutan Survival Foundation. Thank you also to Zaida & Gerry Butters, Bobbi & Jack Teliska, E-len & David Fu, and Anne Munitz for their generous support.
Competing interests
The corresponding author confirms on behalf of all authors that there have been no involvements that might raise the question of bias in the work reported or in the conclusions, implications, or opinions stated.
Ethical statement
None.
Open access
