Ecosystem Decay in Closed Forest Fragments 317et al. 2006). Exclosure experiments revealed that
arthropod herbivory was a major factor in the fail-
ure of seedlings to survive (Lopez and Terborgh
2007). Strikingly, no tree species exhibited positive
population growth on the S or M islands, demon-
stratingthatthetrophiccascadeaffectedtheentire
tree community. These results suggest that the
effectiveness of plant anti-herbivore defenses is
limited in the absence of top-down control.
In all likelihood, multiple effects of isolation
and hyperabundant consumers contributed to
the high mortality and low recruitment of tree
saplings on S and M islands. However, we strongly
suspect that elevated herbivory was the main
driver, particularly herbivory by leaf-cutter ants
since they forage from the ground to the canopy,
whereas iguanas and howler monkeys forage
only in the canopy. A leading role of leaf-cutter
ants is supported by three observations. First, M
islands supported lower densities of leaf-cutter
ants than S islands. Sapling stem numbers were
higher on M islands and mortality rates lower
than on S islands. Consistent with this, we found
a strong negative relationshi pbetween the density
of leaf-cutter ants and sapling recruitment rates
(r^2 =0.58,P<0.05; unpublished data). Second,
on the S islands, mortality exceeded recruitment
for all tree species. It is unlikely that deficien-
cies of pollinators or dispersers would negatively
impact all members of a tree community because
a multiplicity of agents is involved and fragmen-
tation does not impede them all (e.g., wind; also,
hummingbirds were frequent on S islands; Feeley
2003). Third, S islands tended to be dominated
by species carrying foliage that was determined
through independent tests to be unpreferred by
either howler monkeys or leaf-cutter ants (Rao
et al. 2001, Feeley and Terborgh 2005, Orihuela-
Lopezet al. 2005). The most obvious explana-
tion for this bias is the selective mortality of
preferred species under the pressure of intense
herbivory.
Early in the decade over which the research
unfolded, we wondered whether large fragments
would experience the same changes as small frag-
ments, only more slowly (Levin 1992). In general,
we now feel that the answer is “no” because of the
stepwise addition/loss of ecologically important
species with spatial scale. Leaf-cutter ants offer
a good example. On predator-free S islands they
occurred at a mean density of 4.5 mature colonies
per hectare, whereas on M islands, in the presence
of armadillos, colony density was only 0.2 per
hectare. On the mainland, where colonies are
exposed to army ants (Eciton hammatum) as well as
armadillos and giant anteaters (Myrmecophaga tri-
dactyla), colony density dropped to approximately
0.05 per hectare (Vasconcelos and Cherrett 1997,
Terborghet al. 2001). Similarly, rodent hyper-
abundance is likely to disappear at a fragment size
large enough to sustain ocelots, raptors, and/or
snakes. At large enough scales, ungulates will
replace leaf-cutter ants as herbivores and agoutis
and peccaries will complement small rodents as
seed predators (Wrightet al. 2000). Thus, impor-
tant consequences of fragmentation result from
the presence/absence of individual species, a fact
that raises questions about the use of area as
a continuous variable in fragmentation studies
(e.g., Wardleet al. 1997).The broad consequences
of fragmentation – community change accompa-
nied by biodiversity loss – will occur over a wide
range of spatial scales, but the details (e.g., the
mortality and recruitment schedules of individual
species) will certainly differ.
An important question for conservation is
whether there is an upper limit to the scale effects
of fragmentation.Researchonungulatepredator–
prey systems in North America strongly implies
that the presence of top carnivores (wolves or the
equivalent) is necessary to avert top-down trophic
cascades (Mclaren and Peterson 1994, Rippleet al.
2001). The areas needed to sustain populations
of to pcarnivores are in the thousands of square
kilometers (Woodroffe and Ginsberg 1998).
A related question of conservation relevance
is whether the processes observed in closed frag-
ments extend to open fragments in a mainland
land-use matrix. The question has no simple
answer because mainland systems are typically
more complex and less controlled than those on
forested islets. In open mainland systems, non-
resident animals are free to move through the
matrix and resident animals have the option of
density-dependent emigration. Moreover, human
activities such as hunting, roadkill, agricultural
chemicals, domestic animals, fire control (or the
lack of it), and the qualities of the matrix can all