Specialization of plant-pollinator interactions increases with temperature at Mt. Kilimanjaro

Abstract

Aim: Species differ in their degree of specialization when interacting with other species, with significant consequences for the function and robustness of ecosystems. In order to better estimate such consequences, we need to improve our understanding of the spatial patterns and drivers of specialization in interaction networks.

Methods: Here, we used the extensive environmental gradient of Mt. Kilimanjaro (Tanzania, East Africa) to study patterns and drivers of specialization, and robustness of plant-pollinator interactions against simulated species extinction with standardized sampling methods. We studied specialization, network robustness and other network indices of 67 quantitative plant-pollinator networks consisting of 268 observational hours and 4,380 plant-pollinator interactions along a 3.4 km elevational gradient. Using path analysis, we tested whether resource availability, pollinator richness, visitation rates, temperature, and/or area explain average specialization in pollinator communities. We further linked pollinator specialization to different pollinator taxa, and species traits, that is, proboscis length, body size, and species elevational ranges.

Results: We found that specialization decreased with increasing elevation at different levels of biological organization. Among all variables, mean annual temperature was the best predictor of average specialization in pollinator communities. Specialization differed between pollinator taxa, but was not related to pollinator traits. Network robustness against simulated species extinctions of both plants and pollinators was lowest in the most specialized interaction networks, that is, in the lowlands.

Conclusions: Our study uncovers patterns in plant-pollinator specialization along elevational gradients. Mean annual temperature was closely linked to pollinator specialization. Energetic constraints, caused by short activity timeframes in cold highlands, may force ectothermic species to broaden their dietary spectrum. Alternatively or in addition, accelerated evolutionary rates might facilitate the establishment of specialization under warm climates. Despite the mechanisms behind the patterns have yet to be fully resolved, our data suggest that temperature shifts in the course of climate change may destabilize pollination networks by affecting network architecture.

Keywords: altitudinal gradient; climate change; ecological network; functional traits; generalization; mutualistic interactions; network specialization index (H2′); pollination; robustness; specialization.

Figure 1

Change of plant–pollinator specialization along the elevational gradient on Mt. Kilimanjaro at species and network level. (a) Community mean of pollinator specialization (d'), (b) community mean of plant specialization (d') and (c) plant‐pollinator network specialization (H ₂′) decreased with increasing elevation (m a.s.l. = meters above sea level). Dots represent the abundance‐weighted means of species specialization indices (d′) and the H ₂′ values per transect walk. Lines represent predicted relationships derived from linear mixed‐effects models with elevation as single predictor variable and site as a random term. Dot colors indicate the strength of land use intensity

Figure 2

Direct and indirect predictors of mean pollinator specialization on Mt. Kilimanjaro. (a) A priori hypothesized causal structure of the model. Competitive variables within each hypothesis were highlighted with similar colors. Black and colored arrows indicate positive relationship expectations, gray arrows negative relationships. (b) Structure of the full path model after semiautomated preselection of variables. Detailed information on the preselection process are given in the method section. (c) Final path model derived by AIC_C‐based model selection across all possible paths combinations presented in b. Path coefficients and related p‐Values, as well as both marginal and conditional R ² values for all response variables are presented. Dashed lines indicate nonsignificant paths. The presented path model is statistically not distinguishable from a model in which flower richness has no impact on pollinator specialization (ΔAIC_C = 0.05). The global goodness of fit of all path models was estimated with Fisher's C. p‐values > .05 for C indicate that the specific causal structure reflects the data properly. ACT, actual temperature; LUI, land use intensity, MAP, mean annual precipitation; MAT, mean annual temperature; area = habitat area (100 m above and 100 m below the respective study site). All variables were z‐transformed prior to analyses. Statistical details are given in Table S2.3

Figure 3

Impact of taxonomy and species elevational range on pollinator species specialization (d′). (a) Hymenoptera (bees and wasps) were on average more specialized than Diptera (syrphid flies; t = 2.70, p = .010; bees‐ wasps: t = 0.03, p = .74; bees—syrphid flies: t = −2.52, p = .016; syrphid flies—wasps: t = 3.36, p = .002). While black box plots present common summary statistics (with data medians as black line and means as white asterisks), the surrounding violin plots signal (smoothed) probability density of the data at different values. (b) Pollinator specialization was not related to the elevational range size of pollinator species. Dot color in (b) corresponds to the different taxonomic groups, as introduced in (a). We considered only pollinators that we observed at least three times (87 species)

Figure 4

Network robustness against species extinction along the elevational gradient on Mt. Kilimanjaro. Robustness against pollinator extinctions (black triangles) exceeded robustness against plant extinction (white dots); both metrics increased with elevation. Robustness was standardized via null‐model comparison prior to analysis. Lines represent predicted relationships derived from linear mixed‐effects models with elevation as single predictor variable and site as a random term