Temporal dynamics of the hummingbird-plant interaction network of a dry forest in Chamela, Mexico: a 30-year follow-up after two hurricanes

Introduction

Drought and extreme winds are the main natural disturbances in Neotropical dry forests (Durán et al., 2002). Natural factors such as these have been found to affect plant–pollinator interactions. It is likely that natural disturbances such as drought and hurricanes will only continue to intensify in the near future as a result of climate change (Allen et al., 2010; Knutson et al., 2015). Therefore, it is important to investigate the effects of these natural disturbances on biodiversity and ecosystem functioning, including plant-animal interactions.

Interactions among species are the driving force behind the biodiversity of the Earth (Bascompte, 2009) as well as the structure of its communities (Berlow et al., 2009).These interactions can be represented as networks, enabling the topology (or patterns) of the relationships among species to be characterized. Accordingly, these networks provide a simple and powerful framework for understanding species interactions (Poisot, Stouffer & Gravel, 2015). Additionally, network theory has informed several approaches for statistically quantifying and comparing patterns across communities (Bascompte & Jordano, 2007). These approaches are essential for understanding the ecological, evolutionary, and co-evolutionary dynamics of a network that cannot simply be deduced from pairs of interacting species (Bascompte, 2009).

At a basic level, network structure reflects the organization of species interactions according to the frequency of interactions per node (Proulx, Promislow & Phillips, 2005; Bascompte, 2009). Overall, species interactions tend to be highly heterogeneous: Few species have many interactions, and many species have only a few interactions. In particular, plant–pollinator networks have several common features (Bascompte et al., 2003; Lewinsohn et al., 2006; Guimarães et al., 2007) as well as specific ecological characteristics and structure (Bascompte et al., 2003; Lewinsohn et al., 2006; Guimarães et al., 2007; Morales & Vázquez, 2008; Vázquez et al., 2009). For example, plant–pollinator networks tend to be nested: Generalist species mainly interact with other generalists, forming a core, and specialists also mainly interact with generalists and rarely with other specialists (Bascompte et al., 2003). These patterns of interaction can be explained by ecological processes and evolutionary history (Poisot, Stouffer & Gravel, 2015; Díaz-Castelazo et al., 2010).

However, there is a scarcity of research on how networks change over long periods of time, especially after extreme natural events (Dupont et al., 2009; Díaz-Castelazo et al., 2010). The few studies examining temporal changes in species interaction networks have reported that these networks retain basic topological properties even when plant–pollinator interactions vary over time (Alarcón, Waser & Ollerton, 2008). Notably, species interaction networks that show high species turnover and low variation in their structural parameters (Petanidou et al., 2008; Dupont et al., 2009; Luviano et al., 2018) may be less sensitive to disturbance effects as a result of climatic changes and hurricanes.

Natural events have several wide-ranging effects on natural vegetation that may affect animal-plant interaction networks. In dry forests, most plant species are adapted to drought (Meir & Pennington, 2011). However, extreme winds during hurricanes can result in vegetation damage or fallen trees, which create forest gaps (Askins & Ewert, 1991). Also, vegetation may be severely defoliated, and most flowers and fruits may be dropped (Lynch, 1991; Waide, 1991). Such changes to the original canopy structure may further affect the production of flowers, the primary food resource for pollinators (Askins & Ewert, 1991). Following natural events, greater (major) defoliation has been observed in the superior canopy strata compared to the lower canopy strata and understory, where re-growth tends to be faster (Wunderle Jr, Lodge & Waide, 1992; Brokaw & Walker, 1991).

Aside from vegetation damage, a hurricane can kill birds as a result of wind and rain exposure (Wiley & Wunderle, 1993). Surviving birds can be debilitated and subjected to additional mortality or predation (Lugo, 2008). After Hurricanes Gilbert and Hugo, several studies reported that resident nectarivorous birds were among the most affected (Askins & Ewert, 1991; Lynch, 1991; Waide, 1991; Wunderle Jr, Lodge & Waide, 1992). Nonetheless, the subsequent recovery of their populations to previous numbers suggests that reductions in their populations were due to regional movements as individuals searched for food rather than mortality (Waide, 1991). After Hurricane Gilbert in Jamaica Wunderle Jr, Lodge & Waide (1992) found that nectarivorous species diminished their numbers in the affected areas yet increased their numbers in nearby, less affected areas.

With respect to hurricanes, most ecological studies have focused on their impacts on vegetation (Baldwin et al., 1995; Lomascolo & Aide, 2001), animal communities (Covich et al., 1991; Lugo, 2008), or both (Ferguson et al., 1995; Pascarella, 1998; Rathcke, 2001; Angulo-Sandoval et al., 2004; Horvitz, Tuljapurkar & Pascarella, 2005) although, as previously mentioned, few have documented the effects on species interaction networks (Sánchez-Galván, Díaz-Castelazo & Rico-Gray, 2012; Luviano et al., 2018). In one case, Sánchez-Galván, Díaz-Castelazo & Rico-Gray (2012) studied the effect of Hurricane Karl on a plant-ant interaction network in Veracruz, Mexico. These authors found that most changes occurred in the presence and proportions of interacting species. Also, Luviano et al. (2018) documented the impact of Hurricane Jova on plant-herbivore interaction networks along a successional chronosequence in a dry forest of the Chamela region in Jalisco, Mexico. These authors found that some network descriptors were altered; however, the overall network topology associated with forest succession remained unaltered.

We studied herein the hummingbird-plant interactions in a dry forest of Chamela, Mexico. Our aim was to assess temporal changes in (1) the number and identity of interacting species and (2) the topology of the interaction network over a 30-year period during which several disturbances occurred. Considering the results of previous studies (Alarcón, Waser & Ollerton, 2008; Petanidou et al., 2008; Dupont et al., 2009; Díaz-Castelazo et al., 2010; Dupont et al., 2009; Luviano et al., 2018), we expected that the hummingbird-plant interaction network in the dry forest of the Chamela Biological Station would maintain most hummingbird species over time and that greater changes would be observed in plant species (including the generalist core), mainly due to changes in vegetation cover caused by the passage of Hurricanes Jova and Patricia. As a consequence, we also expected changes in the number and identity of some interactions (i.e., more connections between hummingbirds and lower-strata plants such as shrubs, vines, and herbs and fewer connections between hummingbirds and higher-strata plants such as large trees),yet we expected the general network structure to remain unaltered (i.e., that most network parameters would remain unchanged).

Materials and Methods

Results

The networks differed in the number of interacting hummingbird and plant species. The 1985–1986 network included 21 plant and 5 hummingbird species, and the 2016–2017 network included 27 plant and 4 hummingbird species. Both networks had 45 interaction links (Fig. 1). A total of 40 plant species were visited by hummingbirds (Table 1), yet only 8 plant species were shared by both periods. In the 2016–2017 network, 13 plant species previously reported in the 1985–1986 network were absent, and 19 new plant species were added. Some plant families that lost species included Cactaceae, Bromeliaceae, and Fabaceae but, in this latter family, several new species were also added. Conversely, of the 5 hummingbird species registered in the 1985–1986 network, only Heliomaster constantii was absent in the 2016–2017 network. Because of the difficulty of correctly distinguishing Archilochus colubris and A. alexandri females and juveniles under field conditions, both species were conjointly considered in the analyses.

We observed several changes in the core and periphery species of both networks (Fig. 1). Of the hummingbird species in the 1985–1986 network, Amazilia rutila (Gc = 1.25) was the only core species, whereas in the 2016–2017 network, none of the hummingbirds were considered to be core species, even though Cynanthus latirostris (Gc = 0.91) obtained a Gc value near one. Of the plant species in the 1985–1986 network, 9 out of 21 species (42.8%) belonged to the generalist core, whereas in the 2016–2017 network, only 5 out of 27 species (18.5%) belonged to this group. Of the original plant species core, only C. farinosum, I. bracteata, and J. candicans maintained this status in the 2016–2017 network.

Most network parameters were similar in both sampling periods (Table 2). However, robustness in plant species differed (0.88 and 0.93, P = 0.004), demonstrating that plants became 5.6% more robust to extinctions in the second sampling period. Niche overlap in hummingbirds also increased significantly (51.8%) from the first period to the second (0.27 and 0.41, P = 0.03), but we did not observe a difference in plants (0.41 and 0.36, P = 1.0).

Discussion

We found that natural disturbance events such as hurricanes may have impacted some of the structural parameters of the studied hummingbird-plant network in Chamela, Mexico. Plant robustness and hummingbird niche overlap increased after the disturbance events, and the generalist species core of the network was reduced in number. However, overall, the number of species in the interaction network after the disturbance events was larger than in the prior survey because of the inclusion of new plant species, even though many of the plant species present in the first survey were lost. Notably, four out of five hummingbird species were retained in the second survey, with the exception of Heliomaster constantii.

Temporal changes in the number and identity of species forming interaction networks have been previously reported. For example, a three-year study in California (United States) registered changes in species interactions, although few differences were noted in the number of plants and pollinators (Alarcón, Waser & Ollerton, 2008). Dupont et al. (2009) studied pollination networks across different latitudes and found that species identity and interactions changed from one year to the next, although the number of species and interactions remained constant. Petanidou et al. (2008) reported that most species and interactions changed over a four-year period near Athens (Greece): Only 53% of plant species, 21% pollinator species, and 4.9% of interactions remained constant. Díaz-Castelazo et al. (2010) showed that, over a period of 10 years, ant-plant networks in Veracruz (Mexico) differed in the number of species and interactions, suggesting a high species turnover. These findings suggest that the actual range and strength of interactions in the plant and pollinator community vary over time and are unlikely to be detected over short periods of time (Alarcón, Waser & Ollerton, 2008). Long-term studies on plant–pollinator networks are needed, especially after disturbance events such as droughts and hurricanes, as the composition of interacting species can change following these events. Additionally, the modification of plant abundance, flowering, or pollinator phenology as a result of climatic changes can affect the plant–pollinator network (Burkle & Alarcón, 2011).

Two epiphytic plant species were lost in the 2016–2017 interaction network including T. paucifolia, a generalist core species in the 1985–1986 network. Extreme events can cause the direct mortality of epiphytic plants or damage their host trees (Robertson & Platt, 2001). In Chamela’s dry forest, droughts are part of the intra- and interannual dynamics of the region (Martijena & Bullock, 1994; Maass et al., 2017); thus, the inhabiting species are likely well adapted to these events. However, the abundance of epiphytic species was impacted, for example, after Hurricanes Andrew, Hugo, and Wilma in the Caribbean (Migenis & Ackerman, 1993; Loope et al., 1994; Goode & Allen, 2008). As a result of climate change and the corresponding predicted increase in hurricane incidence, epiphytic plants may disproportionally suffer compared to other understory plant species.

Cacti adapted to drought conditions such as Nopalea karwinskiana and Opuntia excelsa also disappeared from the 2016–2017 network. The first cactus was restricted to rocky creek edges yet, after unusually heavy rains in 1998, it completely disappeared (MC Arizmendi, pers. comm., 1998). It was a periphery species but was important for H. constantii (Arizmendi & Ornelas, 1990), which was not registered in the second survey. Many O. excelsa individuals (a generalist core plant species, particularly important for H. constantii and A. rutila hummingbirds) fell or lost their terminal cladodes (those that produce flowers) due to wind force and felled trees after Hurricane Patricia (S Díaz Infante, pers. comm., 2016), as reported for other similar events (Frangi & Lugo, 1991). The effects after Hurricane Jova were probably similar but likely occurred at a lower intensity due to the lower wind force of this hurricane.

Most of the rest of missing plants in the 2016–2017 network had few interactions and were periphery species loosely connected to hummingbirds, so their probability of interaction with hummingbirds was already low (Borgatti & Everett, 2000). Notably, blooming was not observed in tall trees such as Tabebuia donell-smithii and P. mangense during the 2016–2017 survey. These species are more susceptible to damage and mortality as a result of strong winds during hurricanes (Wurman & Winslow, 1998; Segura et al., 2002; Vandecar et al., 2011). Trees such as Erythrina lanata and Ceiba aesculifolia, located on slopes and hills, did not bloom either after Hurricane Patricia, as most of their treetops were removed after the hurricane. This probably also occurred with Hurricane Jova. Parker et al. (2017) found that both hurricanes altered the forest structure in Chamela, but Hurricane Patricia had a larger and more devastating effect on fruit and flower abundance and phenology than Hurricane Jova (Renton et al., 2017).

Most of the new plant species added to the network are mainly pollinated by insects. They had few interactions with hummingbirds, and none belonged to the generalist core. Thus, the probability of detecting their interactions before was also low (Van der Pijl, 1961; Tripp & Manos, 2008). Moreover, most new species were vines or shrubs (8 spp.), which likely benefitted from the newly opened gaps after Hurricane Patricia. For example, even though the upper strata (4 to 11 m) lost up to 83% ± 16% of canopy material, the lower strata (1 to 3 m) increased its cover by 61% ± 68% after the hurricane, mainly due to the growth of vines and shoots (Parker et al., 2017). One unique case is Tabebuia rosea, a tree species damaged by wind but included as a core species in the second survey; most registered individuals were planted after the first survey (JM Verdusco, pers. comm., 1987).

Overall, our data concur with reports in other study sites affected by hurricanes. For example, the number of fruits and flowers on woody plants was abnormally low several months after Hurricane Hugo impacted the Virgin Islands (Askins & Ewert, 1991); Waide1991). Likewise, vines, branches, and invasive herbs formed an almost impenetrable mass in the understory of a tropical forest in Quintana Roo, Mexico, several months after the arrival of Hurricane Gilbert (Lynch, 1991). After Hurricane Hugo in Jamaica, the lower strata of foliage recovered faster than the upper strata because of the more rapid growth of herbaceous plants and the sprouting of trees and shrubs. Even after 17 months, treetops had not recovered their original density due to fallen trees and broken trunks and branches (Wunderle Jr, 1995). Similarly, after Hurricane Gilbert in Puerto Rico, understory growth compensated for the loss of treetops (Wunderle Jr, Lodge & Waide, 1992). However, despite the severe damage to vegetation, the species composition of the studied plant communities was speculated to remain unaltered after Hurricanes Hugo and Gilbert (Brokaw & Walker, 1991). In our study, an important fraction of the species absent in the 2016–2017 network were either susceptible to hurricane damage or had low interaction probabilities. Of the newly added species, many had few interactions or were herbs, vines, or shrubs, which appear to have benefitted from the passage of the hurricanes. According to Olesen et al. (2008), the new species added in a network are usually specialized or often rare.

As with plants, the hummingbird species that were originally more connected (A. rutila, C. latirostris, and C. auriceps) were maintained in the network, whereas the species with fewer interactions (H. constantii ) and those associated with disturbance-vulnerable plant species such as O. excelsa, C. aesculifolia, and N. karwinskiana were absent. Amazilia rutila, the only core species in the first network, was closely associated with O. excelsa, a plant affected by hurricane winds. Previous studies have shown that frugivorous and nectarivorous birds are more susceptible to a decline in food resources than other birds, as occurred on the Virgin Islands with the hummingbird Orthorhynchus cristatus after Hurricane Hugo (Askins & Ewert, 1991) and in the Yucatán Peninsula in Mexico with nectarivorous birds in general after Hurricane Gilbert (Lynch, 1991). Also, rare hummingbird species are more susceptible to disappearance from an affected area, as occurred with the Vervain Hummingbird (Mellisuga minima) in Jamaica after Hurricane Gilbert. However, even abundant hummingbirds, such as the Streamertail Hummingbird (Trochilus polytmus), may suffer considerable mortality from exposure to hurricanes or may wander to nearby less damaged areas(Wunderle Jr, Lodge & Waide, 1992), especially considering that tropical nectarivorous birds have a wide range and often migrate seasonally and altitudinally in some tropical regions (e.g., Stiles, 1983).

On the other hand, the two Archilochus species seemed to benefit from hurricane-borne disturbance in our study. These generalist winter migrant hummingbirds are commonly associated with disturbed areas (Finch, 1991; Udvardy & Farrand, 1994; Arizmendi & Berlanga, 2014) and mainly associated with the vine Ipomoea bracteata. Previously, according to Lynch (1991), overwintering Nearctic migrants appeared to be more resilient than year-round residents to hurricane effects. The latter author found that these migrants, who are generally associated with edges, dense scrub, and open zones, invaded a forest damaged by Hurricane Gilbert in Quintana Roo, Mexico. Thus, it seems that migratory generalist species are favored by disturbance events and the growth and blooming of vines and shrubs.

With respect to the parameters of the hummingbird-plant interaction network, only plant robustness and hummingbird niche overlap presented significant differences after 30 years. Several studies describing temporal patterns in pollination networks revealed similarities (Burkle & Alarcón, 2011). For example, most studied pollination networks have been shown to be highly nested over time (Alarcón, Waser & Ollerton, 2008; Petanidou et al., 2008; Dupont et al., 2009; Díaz-Castelazo et al., 2010). Connectance values also tend to be conserved over time (Alarcón, Waser & Ollerton, 2008; Olesen et al., 2008; Petanidou et al., 2008; Dupont et al., 2009; Díaz-Castelazo et al., 2010). As sampling time increases, specialization tends to diminish (Petanidou et al., 2008) because pollinators may use less preferred floral resources when their favorite plant species are scarce or absent (Burkle & Alarcón, 2011). However, interaction network characteristics appear to remain constant within a framework of annual variation in species identity (Petanidou et al., 2008). In our study, the generalist core of nine plants in 1985–1986 was reduced to five in 2016–2017; only three of the original species were conserved. Also, the only former core hummingbird species (A. rutila) lost this status in 2016–2017, probably due to the disappearance of two core plant species (O. excelsa and C. aesculifolia) that served as resources for this hummingbird, and H. constantii was also lost after the hurricanes. In southern California (United States), the identity of core species also changed in plant–pollinator networks studied over three years (Alarcón, Waser & Ollerton, 2008), indicating that specialists interact with different groups of generalists in different years given that even generalist species are subject to fluctuations in population size.

The fact that mutualistic networks form well-defined and predictable patterns of interdependence supports a community-wide approach for species conservation (Bascompte & Jordano, 2007). Network analyses can contribute to conservation strategies at the community level and can highlight important species for the network that should be conserved and protected (Lara-Rodríguez et al., 2012). In addition, the most relevant species in terms of abundance and interaction frequencies can provide insights into the ecological and evolutionary history of interaction networks (Bascompte & Jordano, 2007). Theoretical evidence concludes that heterogeneous, nested, mutualistic networks confer network robustness to species loss and habitat fragmentation (Memmott, Waser & Price, 2004; Fortuna & Bascompte, 2006; Burgos et al., 2007).

However, overall robustness to random species extinctions may be partly explained by the role of a few highly connected species in a network’s core with which specialists interact (Bascompte, 2009). Thus, the robustness of an entire network depends on these species. In Chamela, I. bracteata, C. farinosum, and J. candicans remained as core species overtime, even after extreme events. Therefore, these species significantly contributed to the connectivity and resilience of the studied plant-hummingbird network (Sánchez-Galván, Díaz-Castelazo & Rico-Gray, 2012). Moreover, an increase in hummingbirds’ niche overlap suggests that generalist pollinators can expand their diet (e.g., Archilochus spp. and C. auriceps) when resources are scarce (Waser, 1986), thereby increasing plant robustness. In conclusion, it appears that, despite the disturbances caused by hurricanes, most plant species survive (Burgos et al., 2007), especially if the remaining pollinators become more generalist.

We also considered other factors that could have caused the observed changes in the interaction networks, even though we did not register plant-hummingbird interactions immediately before the hurricanes. However, to the best of our knowledge, no other natural or human-induced disturbances could be directly related to the observed changes in this well-studied site (besides the ones already discussed). On the other hand, we have previous evidence supporting that hurricanes shape forest structure, influence their species composition and diversity, and regulate their functioning (Lugo, 2008).

As stated before, some severe droughts were recorded at the study site, but plants and animals native to the area are adapted to a highly unpredictable and strongly seasonal rainfall pattern (Maass et al., 2017). Drought can affect plants’ flowering, but its effects in dry forests are usually restricted to the seasons/years when it occurs and the following year/season. Thus, if plants are able to resist drought, there is usually not a lasting effect on flowering. Furthermore, in the last 33 years, there have not been any reported vegetation changes in the area due to drought. Thus, we can assume that the reason that many plants or trees did not bloom during the second survey was hurricane damage rather than drought, especially considering that these effects are not normal in the study site. Finally, we were able to directly observe damage to the vegetation cover that was also reported in the literature (Maass et al., 2017; Jimenez-Rodríguez et al., in press; Parker et al., 2017).

Even so, distinguishing the effects of a large, infrequent event from the background variation of a seasonal ecosystem is challenging (Parker et al., 2017). Thus, we recognize that only longer-term studies would evidence network fluctuations due to normal inter-annual variation in the abundance and phenology of plant and hummingbird populations. Also, a recovery of vegetation and decrease in new opportunistic species would confirm that natural disturbances had been influencing the interaction network.

However, some external factors not considered herein may also contribute to the observed interactions (e.g., external effects on migratory hummingbirds). Either way, if plant–pollinator interactions are dynamic and opportunistic, this should increase the resistance of pollination networks to species loss and phenological changes (Alarcón, Waser & Ollerton, 2008). Thus, even in the face of extreme events, these networks may persist.

Conclusions

The number and identity of interacting species in the hummingbird-plant network of a dry forest in Chamela, Mexico, primarily plant species, changed over a 30-year period. We mostly observed changes in the less connected species and those more prone to be affected (e.g., epiphytes, cacti, and large trees) or benefitted (e.g., herbs, vines, and shrubs) by hurricanes. In the case of hummingbirds, all but one (Heliomaster constantii) remained in the network over time. Overall, most network structural parameters remained unchanged, with the exception of plant robustness and hummingbird niche overlap, which both increased. Also, extreme events appear to have impacted the generalist core of species, resulting in the loss of several species. Finally, the most important plant species for conserving the plant-hummingbird interaction network in the Chamela-Cuixmala Biosphere Reserve are Ipomoea bracteata, Combretum farinosum, and Justicia candicans. The temporal perspective of this study provides unique insights into the conservation of plant-hummingbird networks across time and after extreme natural events.

Supplemental Information