Spatio-temporal effects of climate change on the geographical distribution and flowering phenology of hummingbird-pollinated plants

Abstract

Backgrounds and Aims

Tropical plant species are already suffering the effects of climate change and projections warn of even greater changes in the following decades. Of particular concern are alterations in flowering phenology, given that it is considered a fitness trait, part of plant species ecological niche, with potential cascade effects in plant–pollinator interactions. The aim of the study was to assess the potential impacts of climate change on the geographical distribution and flowering phenology of hummingbird-pollinated plants.

Methods

We implemented ecological niche modelling (ENM) to investigate the potential impacts of different climate change scenarios on the geographical distribution and flowering phenology of 62 hummingbird-pollinated plant species in the Brazilian Atlantic Forest.

Key Results

Distribution models indicate future changes in the climatic suitability of their current habitats, suggesting a tendency towards discontinuity, reduction and spatial displacement. Flowering models indicate that climate can influence species phenology in different ways: some species may experience increased flowering suitability whereas others may suffer decreased suitability.

Conclusions

Our results suggest that hummingbird-pollinated species are prone to changes in their geographical distribution and flowering under different climate scenarios. Such variation may impact the community structure of ecological networks and reproductive success of tropical plants in the near future.

INTRODUCTION

Anthropogenic climate change is increasingly recognized as a major threat to current biodiversity and human well-being, affecting geographical distributions, phenology and even leading to extinction of organisms across a wide range of taxonomic and functional groups (Parmesan and Yohe, 2003; Parmesan, 2006; Zwiener et al., 2018). In plants, responses to climate change can be diverse, including changes in species geographical distribution and phenology (Parmesan and Yohe, 2003; Walther et al., 2005). For instance, increasing temperatures can trigger early leaf opening, flowering and germination, and delay leaf fall, leading to a longer growing season (Cotton, 2003).

Plant phenology responds to climate change in different ways, and while most documented phenological changes support earlier events, some species show delayed responses (Parmesan and Yohe, 2003). Such variation is due to the complexity of the factors that modulate phenological events, such as biotic components, that determine the intensity of the event, and climatic factors (i.e. temperature, precipitation and photoperiod) that affect the period and the duration of a phenological event (Schaik et al., 1993). Thus, the entire life cycle of plants may suffer seasonal changes related to habitat suitability (Visser and Both, 2005). The effects of climate change on the phenology of tropical species is of particular concern because such species exhibit restricted physiological tolerance to climate extremes due to the lack of annual climatic seasonality in their native distribution (Morellato et al., 2000). As such, tropical plants may not adapt to very intense climate changes and are expected to present disproportionate effects compared to species of temperate environments (Colwell et al., 2008; Deutsch et al., 2008).

In addition to changes in phenological patterns, organisms are also altering their geographical distributions in response to climate change (Bykova et al., 2012). Flowering phenology is considered a fitness trait that is part of plant species ecological niche, which is mirrored in their patterns of geographical distribution (Chuine, 2010). Therefore, to assess the impacts of climate change on phenology, approaches that include both spatial and temporal responses are needed. In this sense, the prediction of how future climate changes may affect phenology in different scenarios is especially challenging (Visser and Both, 2005).

Flowering phenology is associated with the reproductive success of plants, as it is directly involved with pollination and the resulting fruit and seed formation (Chuine, 2010). Synchronization of the flowering period with the activity pattern of pollinators is of paramount importance for the maintenance of ecological networks, given that it provides conditions for adequate transport and deposition of pollen, as well as nutritional resources for animals (Junker et al., 2012; Machado, 2012). In fact, plants pollinated by animals may be particularly vulnerable to climate change because plants and pollinators may not respond in similar ways to changes in climate, potentially leading to the disruption of plant–pollinator interactions (Gordo and Sanz, 2010; Bartomeus et al., 2011; Dorji et al., 2012; McKinney et al., 2012). Such desynchronization can be detrimental for pollinator populations and in the long term may have negative effects on the reproduction of plants (Aldridge et al., 2011). Bird pollination is considered to be quite common in tropical ecosystems (Rocca-de-Andrade, 2006), and hummingbirds play an important role in the reproduction of several species of plant in the Neotropics (Zanata et al., 2017). In these forests, plants produce flowers throughout the year, providing food resources thar are essential for the survival of these birds (Rocca and Sazima, 2010).

In the Atlantic Forest, many plant species are endemic (BFG, 2015), and particularly sensitive to disturbances (Rabinowitz, 1981), yet no study has assessed the effects of climate change on phenological flowering patterns. This restricts our understanding of this important driver of the biodiversity crisis and its potential effects on the reproductive success of tropical plants and the availability of food resources for their mutualistic partners, such as hummingbirds. In this study, we implemented an innovative approach to investigate the effects of climatic change on flowering patterns and the geographical distribution of hummingbird-pollinated species in a biodiversity hotspot. Based on monthly flowering records from georeferenced occurrence points we modelled spatial changes in species distributions in association with temporal changes in flowering suitability under different climate change scenarios in the Atlantic Forest.

MATERIAL AND METHODS

Approach to modelling species distributions and phenology

To model the potential geographical distribution of hummingbird-pollinated plants and flowering suitability under current and future climatic scenarios, we used ecological niche modelling (ENM) in a two-stage process. First, we estimated species geographical distributions and, in second step, estimated flowering suitability within the potential distributions for each recorded flowering month (Fig. 1). ENM is an important tool for assessing potential effects of climate change on biodiversity (Skov and Svenning, 2004; Thuiller, 2004; Araújo and Guisan, 2006; Peterson et al., 2011) and has been widely used to estimate species geographical distributions under different scenarios (Böhner and Lehmkuhl, 2005; Beaumont et al., 2008; Faurby and Araújo, 2018). This modelling approach has also been used to assess temporal biological patterns, although less frequently, in a phenological and climate change context (Peterson et al., 2005; Barve et al., 2014, 2015).

Fig. 1.

Two-stage modelling approach used for assessing spatiotemporal variation in species geographical distribution and flowering suitability under different climate change scenarios.

Species and occurrence records

We compiled published studies containing data of plant species that interact with hummingbirds in the Brazilian Atlantic Forest (Supplementary Data Table S1). We searched the Web of Science database for studies published from 1982 to 2006 using the following keywords: hummingbird, rainforest, Atlantic Forest, plant, interaction, pollination. Based on the compiled studies we identified 204 angiosperm species having such interactions.

For these 204 species, we obtained primary georeferenced data from biological collections included in the SpeciesLink and Global Biodiversity Information Facility (GBIF) databases. All species were checked for nomenclature and spelling errors using as reference the Flora do Brasil 2020 Project (2016). Additionally, we adopted the following criteria to increase data quality: (1) records originated only from natural populations; (2) precision in the geographical coordinates; and (3) exclusion of records outside the biome and Brazilian states of natural occurrence based on information from botanical specialist (available at http://floradobrasil.jbrj.gov.br).

To reduce spatial autocorrelation of records and over-representation of the environment (Boria et al., 2014), we removed duplicated records and pairs of points separated by <5 km. This distance was defined according to the density of occurrence records and the dispersion characteristics of the selected species. This was done because records from biological collections usually present bias due to ease of access to sampling sites (Lemes et al., 2011), which in turn may affect model performance (Hijmans, 2012; Boria et al., 2014; Fourcade et al., 2014).

From the occurrence records, we verified sampling dates and selected those that described the presence of flowers or floral structures at the time of sampling. For each species, we recorded the temporal resolution of flowering periods in months. Finally, following these criteria, we selected 62 species (Table 1).

Table 1.

Model evaluation for distribution models using a binomial probability test and flowering monthly models using an AUC test.

Species	Binomial probability results	AUC test results
		Jan	Feb	Mar	Apr	May	Jun	Jul	Aug	Sep	Oct	Nov	Dec
Aechmea bromeliifolia	0.9559	–	–	–	–	–	–	–	0.6990	–	–	–	–
Aechmea lamarchei	0.9130	–	–	–	–	–	–	–	0.7053	–	–	–	–
Aechmea nudicaulis	0.8824	0.7972	–	–	–	–	–	–	–	–	0.7402	0.8146	–
Agarista oleifolia	1	–	–	–	–	–	–	–	–	–	0.8083	–	–
Aphelandra longiflora	0.9737	–	–	0.8540	0.7942	0.7673	0.6004	–	–	–	–	–	–
Billbergia amoena	0.8919	–	–	–	–	0.7845	–	–	–	–	–	–	–
Bomarea edulis	0.9913	0.8257	0.8511	–	–	0.9194	–	0.9810	0.9480	–	–	–	0.9160
Bougainvillea spectabilis	0.8919	–	–	–	–	0.7937	0.6627	–	0.8094	–	–	–	–
Buddleja stachyoides	0.9880	–	–	–	–	–	0.7606	0.8173	0.7575	0.7557	–	–	–
Camptosema scarlatinum	1	–	–	–	–	–	–	–	–	–	0.7837	–	–
Centropogon cornutus	0.9648	–	–	0.5	0.8062	0.9116	0.7808	0.8310	0.8639	0.9130	0.8632	–	–
Cestrum corymbosum	0.9726	–	–	–	–	–	–	–	–	0.7255	0.7652	–	–
Cnidoscolus urens	0.9662	0.8447	0.83691	0.8732	0.8649	0.8202	0.7700	0.8922	0.8551	–	0.8187	0.8759	0.7385
Collaea speciosa	1	–	–	–	0.7884	0.8130	–	–	–	0.8568	–	–	–
Combretum fruticosum	0.9695	0.9536	–	–	–	0.7768	0.6507	0.7606	0.7654	–	–	–	0.9047
Costus arabicus	0.8718	–	0.68481	–	0.7099	–	–	–	–	–	–	–	–
Costus spiralis	0.9773	0.7972	0.8540	0.8932	0.7389	0.8329	–	–	–	–	–	–	0.7133
Eriotheca gracilipes	1	–	–	–	–	0.6534	0.6370	0.7025	0.7239	–	–	–	–
Erythrina falcata	0.9672	–	–	–	–	–	–	–	0.7829	0.8296	0.7218	–	–
Erythrina speciosa	0.9211	–	–	–	–	–	–	0.6481	0.8305	–	–	–	–
Esterhazya splendida	0.9742	0.8365	0.7833	0.8348	0.8206	0.8225	0.8449	0.6339	–	–	–	–	–
Fridericia speciosa	1	0.8561	0.6130	–	–	–	–	–	–	–	–	0.7771	0.7582
Fuchsia regia	0.9821	0.7847	0.7339	0.7717	0.6712	–	–	–	–	–	0.8365	0.7823	–
Gaylussacia brasiliensis	0.9283	0.7877	0.7541	0.8726	0.6512	0.7678	0.6795	0.7550	0.8438	0.7732	0.8600	0.9109	–
Hamelia patens	0.9853	–	–	0.7987	–	–	–	–	–	–	–	–	–
Handroanthus chrysotrichus	1	–	–	–	–	–	–	–	0.8080	–	0.7593	–	–
Heliconia psittacorum	0.9731	0.8327	0.8310	0.8563	0.7781	0.9020	0.7492	0.9027	0.8219	–	0.6216	0.7266	0.8643
Helicteres brevispira	1	–	–	–	0.7805	–	–	–	0.6897	0.8335	0.7577	0.8227	–
Inga edulis	0.9710	0.8462	–	–	–	–	–	–	–	–	0.7071	0.8785	–
Inga marginata	0.9663	0.9260	0.9114	0.8660	–	0.8084	0.7336	0.8739	0.8718	0.9086	0.8176	0.8126	0.7831
Inga sessilis	0.9350	–	–	0.7986	0.7996	0.8377	0.6610	–	–	–	–	–	–
Inga subnuda	0.9239	–	–	–	–	–	–	–	–	0.8028	0.8644	–	–
Ipomoea hederifolia	0.9747	–	–	–	0.8142	0.8171	0.7570	–	–	–	–	–	–
Ipomoea quamoclit	0.9667	–	–	–	0.7036	0.8826	–	–	–	–	–	–	–
Jacaranda puberula	0.8874	–	–	–	–	–	–	–	0.7915	0.8610	0.8277	0.7008	–
Justicia brasiliana	0.9570	0.7638	0.8263	0.7893	0.8900	0.6991	–	–	–	–	0.5929	0.8043	0.6036
Justicia carnea	0.9348	0.8346	–	–	–	–	–	–	–	0.8269	0.6394	0.7150	0.6812
Manettia cordifolia	0.9948	0.8641	0.7746	0.8300	0.7886	0.7787	0.8607	0.8760	0.8473	0.8236	0.8545	0.7516	–
Manettia gracilis	1	–	–	–	0.7925	0.7949	–	–	–	–	–	–	–
Mendoncia velloziana	0.9333	0.6597	–	–	–	–	–	–	–	–	0.9151	0.7796	–
Mutisia coccinea	0.9692	–	–	–	–	–	–	0.7747	–	–	0.7670	–	–
Mutisia speciosa	0.9545	–	–	–	–	–	–	–	–	0.5926	–	–	–
Palicourea rigida	1	0.9166	–	–	–	–	–	–	–	0.7044	0.7980	0.8143	0.7261
Passiflora kermesina	0.9773	–	–	–	–	–	–	–	–	–	–	0.8055	–
Passiflora speciosa	1	–	–	–	–	–	–	–	0.8112	0.7708	0.8465	–	–
Psittacanthus dichroos	0.9450	0.7998	–	–	–	–	–	–	–	–	–	–	–
Psychotria nuda	0.9417	–	–	0.8880	0.8433	–	–	–	–	–	–	–	–
Psychotria pubigera	1	0.8899	–	–	–	–	–	–	–	–	–	–	–
Pyrostegia venusta	0.9932	–	–	–	0.7887	0.857	0.8579	0.8345	0.8642	0.8312	0.8569	0.9159	–
Rubus rosifolius	0.96	–	–	0.6533	–	–	–	0.5995	–	–	–	–	–
Ruellia angustiflora	0.9855	–	–	–	–	–	0.7328	0.6470	0.8334	0.8425	0.8091	–	0.5283
Ruellia brevifolia	0.9559	–	0.8071	0.7427	0.8532	–	–	–	–	0.6291	–	–	–
Sabicea grisea	1	–	–	0.8777	–	–	–	–	–	–	–	–	–
Sacoila lanceolata	0.9917	0.9350	0.8744	–	–	–	–	–	0.9381	0.9381	0.9153	0.8199	–
Sinningia allagophylla	0.9857	0.7605	0.7959	0.7785	–	–	–	–	–	–	0.7668	0.8831	0.8100
Sinningia douglasii	0.9828	–	–	–	–	–	–	–	–	–	0.7651	0.8065	–
Sinningia elatior	1	0.8164	0.7371	0.8093	–	–	–	–	–	–	–	0.6856	0.8134
Stachytarpheta cayennensis	0.9563	0.8414	0.8186	0.8241	0.8282	0.7832	0.8406	0.8294	0.8627	0.8038	0.7626	0.8266	0.8492
Tillandsia geminiflora	0.9512	–	–	–	–	–	–	–	0.8145	0.8021	0.8779	0.8135	–
Tillandsia stricta	0.9588	0.7138	0.8274	0.8031	0.9013	–	–	0.5624	0.7890	0.7520	0.7857	0.7584	0.6748
Tillandsia tenuifolia	0.9686	0.6728	0.6701	0.6894	–	0.6386	–	–	0.8748	0.7786	0.83834	0.8862	–
Varronia multispicata	0.9697	–	–	0.7901	–	–	–	–	–	–	–	–	–

Environmental variables

Environmental variables were obtained from WorldClim for the present and future periods (2050 and 2070) at 5^arc-minute spatial resolution (~9 km at the equator). For the development of species distribution models (Step 1), we selected 19 bioclimatic variables, derived from temperature and precipitation (Hijmans et al., 2005). For the modelling of flowering suitability (Step 2), the selected climatic variables were maximum temperature, minimum temperature, mean precipitation and solar radiation, at monthly resolution (January–December). These variables were selected because precipitation and solar radiation are considered to be strongly related to the induction of flowering in tropical forests (Gunter et al., 2008).

For future climate projections we selected three global circulation models (GCMs): CCSM4 (Community Climate System Model) models, GISS-E2-R (Goddard Institute for Space Studies) and MIROC5 (Model for Interdisciplinary Research on Climate). We associated these GCMs with two representative concentration pathways (RCPs) of radiative forcing: 2.6 W m² (RCP 2.6) and 8.5 W m² (RCP 8.5). These represent a low and high intensity of greenhouse gas emission, respectively (IPCC, 2014). To reduce dimensionality and collinearity of variables for ENM in Stage 1, we applied principal components analysis (PCA) to environmental variables in the present and projected the results to future climatic scenarios. We selected the first six components to use as predictors in ENM that together accounted for 95 % of the total variation.

Ecological niche modelling

To estimate species geographical distributions, we employed Maxent v3.3.3K (Phillips et al., 2006), implemented in the R package dismo (Hijmans et al., 2008), using the default settings to develop the models. We generated distribution models (Step 1) for each climatic scenario and period, i.e. three GCMs (CCSM4, GISS-E2-R and MIROC5) and two RCPs (2.6 and 8.5) for 2050 and 2070, overall 12 models for each species (e.g. CCSM4 - RCP 2.6 - 2050; CCSM4 - RCP 2.6 - 2070; CCSM4 - RCP 8.5 - 2050; CCSM4 - RCP 8.5 - 2070).

For flowering estimates (Step 2), the models were generated monthly according to the chronology of flowering records for each species, and using the corresponding monthly climatic data for the same scenarios and time periods from Step 1. In stage 1, models for each species were calibrated and transferred within training regions representing the dispersal capacity of the species (Barve et al., 2011; Olson et al., 2001). These individual accessible areas were delimited using a minimum convex polygon (Burgmann and Fox, 2003; Oliveira et al., 2016) with a buffer of ~50 km around the points of occurrence. Accessible areas were then adjusted according to ecoregions (Olson et al., 2001), characteristics of the landscape and potential biogeographical barriers, such as rivers and mountain ranges.

To obtain binary predictions we applied a lowest presence threshold discarding the 5 % lowest values of suitability (Peterson et al., 2011). They were then grouped by period (2050 and 2070) and RCP (2.6 and 8.5), considering only the locations of agreement between the projections of the three GCMs.

Finally, to estimate flowering suitability for each species (Stage 2) we used the extent of potential distribution models from Stage 1 to crop monthly environmental data, to calibrate models in the months that flowering was recorded for a given species, and to project suitability values within boundaries of estimated geographical distributions, for each climatic scenario and period. The analyses were carried out initially for the entire distribution range of the species within the Brazilian territory, and then cut in the delimitations of the Atlantic Forest.

We used two indices to illustrate changes in the distribution under climate scenarios, adopting the conceptual framework proposed by Real et al. (2010) and adapted by Kou et al. (2011): Index I (increment in favourability) and Index O (favourability overlap). These indices measure the potential range of species from the attributes of adequacy predicted over time.

$${\displaystyle \begin{array}{l}{\displaystyle I=\frac{n_{f-}{n}_p}{n_p}O=\frac{n_{f{\displaystyle \stackrel{}{\cap }}p}}{n_{f{\displaystyle \stackrel{}{\cup }}p}}}\end{array}}$$

Where n represents the number of cells, subscripts p and f represent present and future distributions, and subscripts $f\stackrel{}{\cap }p$ and $f\stackrel{}{\cup }p$ represent overlapping and combined areas of both future and present distributions, respectively. The I and O indices were calculated with binary models (values 0 and 1) in all climatic scenarios described above.

To identify trends of changes in flowering suitability, we use a delta suitability method, calculating the difference between the sum of the suitability values of the future models and the sum of the suitability values of the present model:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{\Delta }S={\displaystyle \sum ^{}}{S}_f-{\displaystyle \sum ^{}}{S}_p}\end{array}}$$

where S represents the climatic suitability values from flowering models, and subscripts p and f represent present and future scenarios. For this calculation, we use the continuous model output for each species.

Model evaluation

To evaluate the performance of distribution models, we used the binomial probability test with the binary predictions. For presence-only data sets, this test indicates the probability that the total number of records correctly predicted by the model can be achieved by a random model (Anderson et al., 2003; Peterson et al., 2011). For each species we partitioned the occurrence records creating a random set of test data with 30 % of the original data. The remaining 70 % were used to train the models, which were later contrasted with the test data set to calculate the probability of presence and absence.

To evaluate phenological models, we used the area under the curve (AUC) of the receiver operating characteristic (ROC) (Fielding and Bell, 1997; Peterson et al., 2011). This is a threshold-independent test that compares the values of suitability with the observed data. Values range from 0 to 1, where values from 0 to 0.5 are interpreted as no predictive ability, and 1 means perfect predictive ability (Araújo et al., 2005; Randin et al., 2006). Here, AUC performance was interpreted using Swets (1988) as adapted by Araújo et al. (2005): >0.90 = excellent; 0.89> <0.80 = good; 0.79> <0.70 = adjusted; 0.69> <0.60 = poor; <0.60 = failure.

RESULTS

The geographical distribution models showed a better predictive performance (>0.85) than a random prediction for all scenarios according to the results of binomial probability tests (Table 1). These models indicate that hummingbird-pollinated species may face changes in the climatic suitability range (Fig. 2).

Fig. 2.

Potential distribution of hummingbird-pollinated species in the Brazilian Atlantic Forest under different climatic scenarios (RCP 2.6 and 8.5) and periods (2050 and 2070).

The greatest impact is seen in the most recent scenarios of low greenhouse gas concentration (RCP 2.6). For all future projections, the breadth of climate suitability results in losses in connectivity, increases in the western direction of the biome, creating a gap to the south and south-east, and a reduction to the north-east.

The rates of potential range shifts corroborate the observed changes in spatial models (Fig. 3). For 62 species validated, we observed that by 2050, the RCP 2.6 scenario (low emission) is more extreme, where 89 % of the species lose distribution area, compared to 56 % in the RCP 8.5 scenario (high emission). In addition, in other scenarios and periods the maximum loss expected is 62 %. By 2050, in the RCP 2.6 scenario about 11 % of species are expected to increase their distribution whereas in the RCP 8.5 scenario the increases are expected for ~44 % of species. By 2070, the number of species losing distribution area is still higher in the RCP 2.6 scenario (62 %) than in the RCP 8.5 scenario (49 %), but the ratio of species with increasing distribution increases under both emission scenarios (38 % and 51 %, respectively) compared to 2050.

Fig. 3.

Increase in range shifts ratio for hummingbird-pollinated species in the Brazilian Atlantic Forest under different climate scenarios (RCP 2.6 and 8.5) and periods (2050 and 2070) using Index I, which measures an increase in favourability. The y-axis gives the percentage of species that are expected to gain and lose in climatically suitable areas under future climate scenarios (x-axis).

With regard to the maintenance of distribution areas in the different future scenarios (Fig. 4), we observed that most species may be impacted with low overlap. In all climate scenarios, the future distribution areas are predicted to overlap by 25–50 %. Depending on the scenario, 10 % to 52 % of hummingbird-pollinated plants may experience a strong reduction in the overlap between current and future distribution or may not overlap at all (Fig. 4).

Fig. 4.

Overlap range shifts for hummingbird-pollinated species in the Brazilian Atlantic Forest under different climate scenarios (RCP 2.6 and 8.5) and periods (2050 and 2070) using Index O, which measures favourability overlap. Overlap ratios represent the proportion of current distribution areas maintained under future climate scenarios.

Within the projected distribution areas, AUC metrics used to evaluate the flowering models varied among species (Table 1). Differences in monthly climatic suitability for flowering between current and future scenarios have shown that species may respond differently to climate changes. We observed that, in general, more species are experiencing a reduction in climatic suitability for flowering than an increase or maintenance of flowering periods (Fig. 5). This difference is greater between August and November where 61–91 % of the species flowering in this period may not have a suitable climate for flowering in the future. Again, in the RCP 2.6 scenario reductions in climate suitability are more evident than in the high concentration scenario (RCP 8.5).

Fig. 5.

Delta suitability for flowering models for hummingbird-pollinated species in the Brazilian Atlantic Forest under different climate scenarios (RCP 2.6 and 8.5) and periods (2050 and 2070). The y-axis gives the months for which flowering data are recorded in the database, while the x-axis gives the number of species that may experience an increase or decrease (reduction) in climatic suitability for flowering under future climate scenarios.

DISCUSSION

In general, ENM estimates of geographical distribution and flowering suitability show that hummingbird-pollinated species are likely to undergo changes in the future, which may influence their occurrence and reproductive success in different ways. We observed that the flowering period of these plants may increase or decrease as a function of climatic suitability within the duration of their currently flowering period.

We also anticipate that variations in future flowering suitability may stimulate phenological breakthroughs or delays, or even limit current flowering patterns from longer to shorter periods. These different responses are due to the sensitivity of flowering phenology to environmental signals (Miller-Rushing et al., 2010) and the different ways in which plants respond to these signals, advancing their phenology as temperature increases, delaying it when precipitation increases, at least for plants in temperate areas (Hufft et al., 2018). Various hummingbird-pollinated plant species may not respond identically under climate change scenarios because shifts in the entire phenological distributions (first, peak, last flowering, and flowering duration) may vary among species (CaraDonna et al., 2014). This is of key importance for plant–hummingbird interactions given that the length of the flowering season strongly affects the number of plant-interacting partners (Olesen et al., 2008). In addition, climate-driven changes to floral abundance across flowering seasons have been reported to affect pollinator populations due to extensions on flowering length, but also due to the emergence of resource gaps during the season (Ogilvie et al., 2017).

We did not measure flowering intensity, but the constraint on monthly climate suitability could potentially influence changes in flowering patterns, leading to the production of a similar quantity of flowers but in a shorter period, particularly for species with longer blooms, thus affecting resource availability throughout the year. In fact, time and intensity are aspects of flowering phenology that can vary independently of one another, resulting in changes in flower abundance at a population level (CaraDonna et al., 2014; Høye et al., 2013; Ogilvie et al., 2017).

In addition to isolated effects of climate change on geographical distributions and phenology, we showed that variation in flowering suitability can result from the interaction with geographical changes. Such complex effects may arise due to differentiated plant responses to their climatic niche axes. In fact, plants can change their distribution by tracking spatial changes in their climatic niche, whether due to variation in temperature (Le Roux and McGeoch, 2008) or precipitation (Crimmins et al., 2011), as already reported for flowering phenology.

The observed reduction in range followed by expansion in future climate scenarios for hummingbird-pollinated species may result from interlinked changes in temperature and precipitation with unexpected results. For example, differential changes in water availability at different altitudes resulted in unexpected downward shifts for montane plants (Crimmins et al., 2011). However, ENMs assume that species are in equilibrium with the environment where they occur; limitations in characterizing species responses to environmental variation across their geographical range may add uncertainty to model outputs.

We still do not know how climate change will affect the spatiotemporal distribution of hummingbirds. We might expect that the responses of plants and hummingbirds may differ and potentially not match (McKinney et al., 2012). Given that upslope shifts are predicted for many plant species, we may see interaction uncoupling or even a shortfall of pollinators because some elevations might not be suitable for their current pollinators. Indeed, declines in wing-loading capacity of hummingbirds along an elevational gradient have been reported (Feinsinger et al., 1979), with some flight performance traits limited at higher altitudes (Altshuler et al., 2004). In addition, it is also known that in the Atlantic Forest, the composition of pollinator communities changes across the elevational gradient, with a greater representation of some clades at lowlands (Buzato et al., 2000; Wolowski et al., 2017). These to lines of evidence point to a change in the composition of pollinators in future climate scenarios with potential implications for ecological interactions.

The predicted changes in the spatiotemporal patterns of flowering phenology for hummingbird-pollinated species may pose a threat to interactions because pollinator communities are known to be structured by spatial and temporal variations in the diversity of floral resources (Olesen et al., 2008; Burkle and Alarcon, 2011; Carstensen et al., 2014). Phenological differences between trophic levels and interacting partners have been reported in many systems, with increased asynchrony among the interacting species (Parmesan and Hanley, 2015). These independent changes in the multiple facets of flowering phenology may have different consequences for plant reproduction and establishment of interactions (CaraDonna et al., 2014; Fagan et al., 2014; Ramos Jiliberto et al., 2018). In addition, climatic stability is a key factor regulating the availability of food for hummingbirds and indirectly the richness and composition of hummingbird communities (Abrahamczyk and Kessler, 2015).

In this sense, our results highlight potential complete reshaping of plant–hummingbird interactions. However, uncertainty must be considered, given that plant–pollinator networks show strong plasticity (Burkle and Alarcón, 2011) and that the process of rewiring (i.e. change in the identity of interacting partners) at small temporal and spatial scales may rescue plant reproduction and food resources for animals under future climate scenarios (Carstensen et al., 2014; Simanonok and Burkle, 2014; CaraDonna et al., 2017). The fact that many species are losing phenological synchrony (Donnelly et al., 2011; McKinney et al., 2012) further emphasizes the need to mitigate climate change, as its effects may be detrimental to species persistence and therefore to biodiversity (Visser and Both, 2005), particularly for species with specialized interactions (Memmott et al., 2007).

Despite the limitations inherent to estimating geographical distributions and phenological niches with ENM, here we present an alternative approach to study the consequences of climate change on phenological processes. The changes in flowering patterns documented in various studies compare previous and present periods (Fitter and Fitter, 2002; Parmesan, 2006; McKinney et al., 2012; Chuine, 2010). However, information is still lacking for future scenarios, and our proposal is precisely aimed at filling this knowledge gap, and serving as a basis for other studies. Our results highlight the need for a spatiotemporal view of climate change effects on ecological networks and stress the potential threats to tropical biodiversity.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: Compilation records of hummingbird-pollinated plant species in the Brazilian Atlantic Forest.

FUNDING

This work was supported by CAPES – Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Ministério da Educação do Brasil, scholarship to A.P.A.C.L.) and CNPq – Conselho Nacional de Desenvolvimento Científico e Tecnológico (Ministério da Ciência, Tecnologia, Inovações e Comunicações do Brasil) (MCTI/CNPQ/Universal# 445405/2014–7; PQ scholarships # 309453/2013–5, 313801/2017-7 to I.G.V.).