Skip to main content
Dryad logo

Climate change reshapes the eco-evolutionary dynamics of a Neotropical seed dispersal system


Sales, Lilian et al. (2022), Climate change reshapes the eco-evolutionary dynamics of a Neotropical seed dispersal system, Dryad, Dataset,


Aim: Global changes will redistribute biodiversity, reshaping ecological interactions and ecosystem processes. The distribution decoupling of plants and their mutualistic seed dispersers, for instance, may have overlooked eco-evolutionary effects. How animal-dispersed plants will respond to changes in the distribution of their seed dispersers is, however, an open question. Here, we forecast the consequences of climate change and frugivory interactions for the spatial distribution and seed size evolution of a Neotropical palm.

Location: Atlantic forests of South America

Time period: Present day, end of 21st Century

Major taxa studied: Thirty-two species of frugivorous birds, and a palm (Euterpe edulis). 

Methods: Future patterns of animal-plant co-occurrence were derived from ecological niche models, climate forecasts, projections of future forest loss, and seed dispersal simulations. We further explored the evolutionary effect of the spatial reorganization of interactions by modelling palm seed sizes as function of changes in the distribution of frugivore traits.

Results: Our models indicate that future climate change and deforestation may reduce the palm’s suitable distribution by 20-50%. However, our simulations suggest that 66% of all remaining future suitable distribution (76.200 km²) would still be inaccessible to the palm without the active dispersal of seeds by frugivores. In addition, novel frugivore communities are projected to have smaller body mass and gape size (-23% and -10%, respectively), due to the loss of large frugivores, which may translate into a 6–17% reduction of seed sizes across the palm’s remaining distribution.

Main conclusion: Our projections indicate that frugivore seed dispersal may be critical to allow occupancy of future habitat by the studied plant. However, loss of large frugivores may affect trait selection regimes, creating hotspots of plant evolution towards smaller seeds. We argue that such complex dynamics from species-specific responses to global change may drive the distribution and evolution of several interacting partners worldwide.


Species occurrence

The occurrence records for 32 frugivore bird species (Supplementary Fig. S1a and Fig. S2) considered to be effective and legitimate seed dispersers of the palm Euterpe edulis (Pizo & Simão, 2001; De Barros Leite et al., 2012; Galetti et al., 2013) were obtained from ATLANTIC BIRDS, the largest dataset on Neotropical bird distribution (Hasui et al., 2018). We further complemented this dataset with occurrence records from the following online databases: SpeciesLink (, Global Biodiversity Information Facility (GBIF;, VertNet (, Biodiversity Information Serving Our Nation (BISON;, Berkeley Ecoinformatics Engine (Ecoengine;, and iNaturalist ( For all species, we followed the IUCN latest consensus on taxonomy and nomenclature of birds (date of search: October 2018). Online databases were downloaded using the function occ() and then collapsed into the same string using the function fixnames(), both from the R package spocc (Scott et al., 2016). Occurrences for E. edulis were complemented with information obtained from the National Center for Flora Conservation (Centro Nacional de Conservação da Flora - in portuguese, CNCFlora;

All occurrences were thoroughly assessed and quality-checked for their completeness and reliability. Occurrence records located exactly on centroids of municipalities and political polygons were removed from the dataset, as well as duplicates, incomplete coordinates, and those from museums and herbaria, using the suite of coord() functions from R package scrubr (Chamberlain, 2016). We cross-checked all bird records and the known extent of occurrence using the IUCN range maps (IUCN, 2019 Downloaded on 09 February 2019). Occurrences further than 200km from the border of species’ IUCN range maps were removed from the dataset. This conservative data-cleaning approach allowed us to only keep reliable occurrence records. To minimize spatial autocorrelation, we further thinned occurrences less than 0.5km from each other, using the function thin() from the R package spThin (Aiello-Lammens et al., 2015). By doing so, we removed a few records to reduce the effect of clustered sampling, while simultaneously retaining sufficient information for modelling species distributions (Aiello-Lammens et al., 2015). All occurrences are available in the Dryad Digital Repository.

Frugivore community redistribution

We modelled the future redistribution of the palm E. edulis and its frugivore community from species-specific distributional data (Moritz et al., 2008), taken from diverse datasets of Neotropical frugivore distributions (sources provided in Supplementary Information). Occurrences were thoroughly checked and had their completeness and reliability quality-checked. After data cleaning, we ended up with a total of 14,531 records, with an average of 435 ± 267 occurrences per frugivore species (Supplementary Table 1). The thrush Turdus rufiventris was the species with the largest number of records (1,146 records) and the guan Crax blumenbachii was the species with the fewest records (42 records). The clean dataset of E. edulis comprised 575 occurrence records (all occurrences are given in supplementary file Supp_occur_R1.csv).

We established the relationship between occurrences and environmental conditions using ensembles of ecological niche models (ENMs) (Araújo et al., 2019) and by projecting realized niches into environmental forecasts (Peterson et al., 2018) (Supplementary Fig. S1a-f). The ENMs were calibrated and evaluated using present-day occurrences of species, modeled as a function of gridded cell-based information of climate (Fick & Hijmans, 2017) and edaphic features (palm models only) (Hengl et al., 2017) (predictor used for each species are given in Supplementary Table S3). Following the relevant standards (Araújo et al., 2019), we used four commonly used methods, namely random forests (RF) (Breiman, 2001), boosted regression trees (BRT) (Friedman, 2001), bioclim (Busby, 1991), and MaxLike (Royle et al., 2012), implemented in the sdm R package (Naimi & Araújo, 2016). Details on model parameterization can be found in supporting information, section Ecological niche models.

Future climate forecasts were based on the 5th IPCC assessment using five climate models (CESM1-BGC, MPI-ESM-MR, MIROC5, IPSL-CM5A-MR, INMCM4) which represent uncertainty in future climate projections. For each climate model, we considered two extreme scenarios of climate change, obtained as representative concentration pathways, or expectations of greenhouse gas emissions from anthropogenic actions. We considered the rcp45 as an “optimistic” Mitigation scenario where emission rates are expected to slow by the year 2030, and the rcp85 as a “pessimistic” or baseline Business-as-usual scenario, according to historical trends of emission rates without additional future efforts to constrain emissions (IPCC, 2014).

Although climate is a strong driver of species distribution, other non-climatic landscape features affect species survival in human-dominated sites. Here, we simulated the effect of thresholds of forest loss on the persistence and movement of frugivorous birds, using ecological information from multiple sources (Stotz et al., 1996; Hoyo et al., 2015; Vale et al., 2018; IUCN, 2019). To simulate species persistence in fragmented landscapes (Supplementary Fig. S1g-k), we superimposed species-specific thresholds of forest cover (Melo et al., 2018) onto climate suitability surfaces. Cells with unfavorable climate conditions and/or cells with less than the minimum forest cover required for species persistence were both attributed as “unsuitable”. By doing so, our definition of “suitable” cell includes not only climate but also landscape attributes known to affect species persistence on human-dominated sites.  Accessibility to such suitable areas (suitable climate and minimum forest cover) was then modeled as the probability of colonization given the landscape matrix and dispersal constraints (Boesing et al., 2018) (details in Supporting Information, sections Forest loss thresholds and Dispersal-restricted potential distribution of frugivores).

Seed dispersal and occupancy of future climate analogs

We simulated the effects of frugivore seed dispersal on the distribution of the palm using a cellular automata dispersal model (Engler & Guisan, 2009; Engler et al., 2012). To investigate the effect of seed dispersal on forecasts of the palm distribution, we compared scenarios of palm occupancy in the absence of frugivores ─ Passive dispersal, to the dispersal of palm seeds by animals ─ Frugivore-based dispersal. We assume that under passive seed dispersal at the local scale there is no dispersal at the geographic scale. For both the palm and the frugivores, occupancy of a target cell in this model in time t+1 depends on its environmental suitability (projected climate suitability and forest cover thresholds), the distance from a source cell in time t (as a function of species-specific dispersal kernels), and the structure of the surrounding landscape (reflected as barriers to dispersal) (Engler & Guisan, 2009; Engler et al., 2012). Colonization of suitable environments by the palm was allowed if (i) environmental conditions became suitable for the palm and the frugivore, and (ii) the target cell was within reach of a potential source cell, given the dispersal ability of the frugivore. That process was repeated considering pairwise palm-frugivore interactions of the 32 frugivore species and the palm. Barrier cells (e.g. non-forest habitats) were incorporated to simulate species-specific landscape resistance to animal dispersal, as the frugivore is assumed to be the movement agent for palm seed dispersal events (see section Palm seed dispersal scenarios, in Supporting Information).

Usage Notes

For each raster file, initial terms indicate species scientific binomials and final terms indicate scenarios of environmental change. For example:

Aramides_cajaneus_rcp45 (species: Aramides_cajaneus, scenario: rcp45)

Note that the final map has 4 categories

0: always unsuitable

1: climate refugia - always suitable

2: potential colonization (became suitable and is accessible)

3: dispersal limitation (became suitable but is unaccessible)

4: non-analog climate (became unsuitable)


Rasters indicating two species contain spatial information on potential co-occurrence, or where environmental conditions are expected to become suitable and are accessible for both species. For example:

Euterpe_Aramides_cajaneus_rcp45 (species: Euterpe [edulis] + Aramides_cajaneus, scenario: rcp45)

Note that the final map has only two categories:

0: unsuitable and/or unaccessible for at least one species of the pair

1: suitable and accessibe for both species (proxy of potential co-occurrence)


Coordenação de Aperfeiçoamento de Pessoal de Nível Superior, Award: 001

Conselho Nacional de Desenvolvimento Científico e Tecnológico, Award: Biota

Universiteit van Amsterdam, Award: Starting Grant