shaped by coevolution among species with long-
term shared histories ( 27 ). To test our ability to
predict novel interactions, we used interac-
tions involving introduced species as a proxy
for interactions in future novel communities.
We trained a species interaction model using
only data on interactions among species native
to the local network study location, then validated
this model with a test set of data involving
the introduced species at each study location.
Using“current”observations to predict“novel”
interactions in this way (Fig. 2), we found that
the accuracy of novel interaction predictions
equaled or outperformed the cross-validated
prediction accuracy of native species inter-
actions (table S1), which demonstrates a strong
ability to predict novel network assembly based
on standard species traits. The differences that
did exist between predicted and observed inter-
actions may be explained in part by incomplete
sampling of empirical networks ( 28 ). Indeed,
many interactions that were predicted to oc-
cur but were not recorded in the database are
known to occur from other literature and
natural history observations (Fig. 2, I to J). The
high capacity to predict“novel”interactions
indicates that species lacking a shared phylo-
genetic history can interact readily and pre-
dictably using easily quantified traits. Because
the same trait-matching processes that operate
in native communities predict which species
interact in novel communities, more subtle
factors shaped by coevolution at long temporal
scales are unlikely to heavily shape network
assembly—and thus limit plant range shifts—
under climate change.
Our third aim was to assess how alteration
of mammal and bird species composition has
affected seed dispersal function globally. We
focused on long-distance dispersal because
of its key role in population expansion, gene
flow, and recovery from landscape disturbance
( 2 ). We calculated an index of long-distance
dispersal that estimates the quantity of seeds
that a given assemblage of birds and mammals
disperse to distances >1 km and that germi-
nate ( 21 ). These estimates are based on models
capturing the probability that each bird and
mammal species participates as a seed dis-
perser, the probability that it disperses a
focal plant species, the quantity of seeds it
disperses, its impact on germination through
gut passage, and the distances it disperses
seeds (fig. S3 and tables S1 and S3). We
developed maps of long-distance dispersal
under four scenarios of present-day bird and
mammal species composition to assess how
current long-distance dispersal function has
been affected by defaunation and invasion
and how it is threatened by species endan-
germent. First, we used current range maps
that reflect the natural and introduced rangesof extant species today (current scenario). Sec-
ond, we considered a scenario representing
mammal and bird ranges as they would occur
today had they been unaffected by extinctions,
range contractions, or introductions (natural
scenario). This represents a counterfactual
present-day scenario of species composition,
with natural ranges constructed using occur-
rence records from recent history or by using
fossil occurrence data and accounting for range
shifts due to environmental changes (21). Third,
we considered a scenario of the extinction of
bird and mammal species listed as vulnerable
or endangered by the International Union for
Conservation of Nature Red List (endangered
species extinction scenario). Fourth, we con-
sidered a scenario representing the extirpation
of species from their introduced ranges (extir-
pation of introduced species scenario). We
calculated the index of long-distance dispersal
for a representative plant species with median
trait values provided by the animal species
present in each grid cell under each scenario.
We found that current seed dispersal func-
tion has steeply declined from its natural level
(Fig. 3, A and B), with declines particularly
widespread outside the tropics. The few re-
gions where current long-distance dispersal by
birds and mammals exceeds the natural level
are primarily island systems with few na-
tive mammal species (Fig. 3B). In a scenario
featuring the loss of vulnerable and endangered
species (Fig. 3C), the greatest effect relative
to current levels would occur in regions in-
cluding Southeast Asia and Madagascar. This
indicates that remaining animal-mediated seed
dispersal function in these regions is largely
provided by species that are already threatened
or endangered. Extirpation of introduced spe-
cies would reduce long-distance dispersal, par-
ticularly on islands (Fig. 3D). To understand
how the change in dispersal function com-
pares with the change in species richness, we
calculated average values within ecoregions
represented by at least 10 grid cells for the
current and natural scenarios (table S4). We
found that the magnitude of dispersal function
loss far exceeds the magnitude of species
richness loss (Fig. 3E). Ecoregions have dis-
proportionately lost the species that provide
most seed dispersal function, particularly large
mammals that move over large distances ( 29 ).
Further, the magnitude of function decline was
also greater than that of functional diversity,
measured using functional dispersion (Fig. 3F),
albeit less so than when compared with rich-
ness. This shows that functional diversity re-
flects function better than species richness, but
functional diversity can still underestimate par-
ticular functions such as long-distance seed
dispersal. Large vertebrates are especially
important for long-distance seed dispersal, and
human activities disproportionately affect these
key seed dispersers ( 8 , 9 ).212 14 JANUARY 2022•VOL 375 ISSUE 6577 science.orgSCIENCE
0.0 1.5 3.0 4.5 6.0NaturalEndangered
spp extinctionExtirpation of
introd. sppBCDCurrentALong-distance dispersal index (>1km)E1:1 lineFanimal spp. r functional diversityFig. 3. Impacts of human activities on long-distance seed dispersal by birds and mammals.
(A) Spatial variation in the long-distance dispersal index. (BtoD) Percent change in dispersal, relative
to current estimates, for the natural scenario, the extinction of threatened and endangered species,
and the extirpation of introduced species. (EandF) Change in long-distance dispersal function versus
change in species richness and functional diversity in the current scenario relative to the natural scenario,
averaged within ecoregions represented by at least 10 grid cells.
RESEARCH | REPORTS