Ecosystem Decay in Closed Forest Fragments 313consumersandlesservertebratestobehyperabun-
dant only on S and M islands that lacked predators
of vertebrates. In contrast, we did not observe
increases in the abundance of these same species
on somewhat larger islands that held populations
of mesopredators (ocelot, raptors, snakes, etc.).
A more direct test of top-down regulation was
conducted by taking advantage of the contrast-
ing densities of leaf-cutter ants on S islands
(4.5 mature colonies per hectare) versus M
islands (0.2 mature colonies per hectare;Terborgh
et al. 2001). Leaf-cutter density was negatively
associated with the presence of armadillos, a
major predator: none of the S islands supported
armadillos whereas all M islands did. Cages placed
over the entrances to youngAttacolonies on
M islands greatly prolonged their survival, but
had no effect on S islands (Rao 2000). Where
armadillos were present, incipientAttacolonies
weresystematicallydugoutanddestroyed.Telltale
claw marks left no doubt that the predators
were armadillos. These results strongly suggest
that the loss of predators caused the extreme
hyperabundance ofAttacolonies on small islands
(Rao 2000).
An alternative hypothesis sometimes evoked
to explain the hyperabundance of animals on
islands is “ecological release” or “density com-
pensation.” These terms refer to increases in
the mean abundance of members of a guild
on species-poor islands in comparison with the
nearby mainland (Crowell 1962, MacArthuret al.
1972, Yeaton and Cody 1974). Density compen-
sation could potentially result from a number of
mechanisms, but is most frequently attributed
to the utilization of novel resources or habitats
in the absence of competitors (Caseet al. 1979,
Wright1980,VassalloandRice1982,Cody1983,
Anjos and Bocon 1999, Rodda and Dean-Bradley
2002).
Density compensation is an unlikely explana-
tion for the hyperabundance of many species at
Lago Guri since their numbers far exceed those
that could be expected in compensation of missing
potential competitors. For example, the biomass
of howler monkeys on some islands was equiv-
alent to almost 4000 kg km−^2 , more than the
total biomass of any known primate commu-
nity in the world (Terborgh 1983, Peres 1997,
Peres and Dolman 2000). Such extreme “den-
sityovercompensation” cannot be explained by
the same mechanisms as density compensation.
Even with niche expansions and niche shifts,
compensatory increases are not expected to sur-
pass mainland density levels (Diamond 1970,
MacArthuret al. 1972). Incomplete compensa-
tion is predicted because resource use efficiency
is expected to be reduced in novel or “inappro-
priate” habitats/food resources (Diamond 1970,
Conneret al. 2000, Rodda and Dean-Bradley
2002).Effects at the producer levelThe presence of hyperabundant consumer pop-
ulations on S and M islands implied that the
vegetation of such islands would be under intense
top-down pressure (Hairston et al. 1960). In
1997, we found that the density of small saplings
on S islands was only one third of those on L
landmasses, in accord with the expectation of
strong negative top-down effects. In a recensus
conducted 5 years later, the densities of small
saplings had decreased further to less than a
quarter of the density on L landmasses. On M
islands, sapling densities in 1997 were indistin-
guishable from those found on L landmasses but
by 2002 had declined modestly (Terborghet al.
2006).
The vegetation of S islands already had a con-
spicuously battered appearance by 1997 when
vegetation monitoring began. By the end of the
5-year monitoring period in 2002, standing dead
trees were sprinkled through the canopies of these
islands, dead branches and vines littered the forest
floor, and the understory was empty. The recruit-
ment of small saplings (stems≥1 m tall and
<1 cm dbh) on S islands was only 20% of that
observed on L landmasses, whereas stems of all
size classes experienced greater mortality than on
large landmasses. Sapling recruitment was sim-
ilarly depressed to 20% of control values on M
islands, but in keeping with reduced herbivore
densities on this class of islands, the decline in
sapling stem numbers was retarded by about a
decade in comparison with S islands (Terborgh
et al. 2006).