314 Jo hn Terborg hand Kennet hFeeley
Multiple pathways to vegetation
decline
Although damaging levels of herbivory, especially
by leaf-cutter ants, appear to offer the most likely
explanation for the low recruitment and high
mortality of saplings on S islands, other processes
could have been involved. For example, decreased
abundance or diversity of pollinators could lower
seed set (Aizen and Feinsinger 1994); lack of seed
dispersers could prevent adequate seed distribu-
tion (Leighet al. 1993, Asquithet al. 1997, 1999,
Asquith and Mejia-Chang 2005); hyperabundant
seed predators such as small rodents could reduce
seed survivorshi p(Asquithet al.1997, Lopez
and Terborgh 2007, Asquith and Mejia-Chang
2005); and/or high herbivory on adult trees could
result in reduced allocation of resources to repro-
duction (Ruess and McNaughton 1984, Belsky
1986). Distortions in any or all of these processes
could contribute to suppressed recruitment of tree
saplings.
Pollination
Despite a steady cross-water flow of immigrant
butterflies (a representative pollinator group) to
many S and M islands, the overall density of
butterflies on these islands was low relative to L
landmasses (Shahabuddin and Terborgh 1999).
Immigrant or experimentally introduced butter-
flies left most S islands within 24 hours. These
islands often lacked either adult food resources
and/or larval food plants (Shahabuddin and Ter-
borgh 1999). Similarly, dung beetles emigrated
quickly from small islands (Larsenet al. 2005).
Based on these results, we surmise that many
flying insects do not turn back when they fly
out over open water, so that individuals tend to
emigrate from smaller islands faster than they
immigrate. Fragmentary evidence (unpublished)
suggests that the reproduction of some tree species
on S and M islands may have been hindered by a
deficiency of pollinators.
Seedpredation
We tested the hypothesis that S and M islands
would have higher rates of seed predation due to
high rodent abundances by setting out arrays of
16 species of seeds on S, M, and L landmasses
and monitoring removal rates. Contrary to expec-
tation, removal rates did not differ between large
landmasses and islands. This seemingly paradoxi-
cal result held for seeds set out on the forest floor as
well as for lightly buried seeds (Lopez andTerborgh
2007). Apparently the species of rodents able to
persist on S and M islands were inefficient seed
predators. Abnormally high rates of seed preda-
tion thus did not seem to be contributing to low
sapling recruitment rates.Herbivory
Finally, we tested the role of herbivory in repress-
ingrecruitmentratesbysettingoutseedlingsof six
species of common forest trees under three types
of cages: impermeable (closed to arthropods and
all larger animals), skirted (open only to arthro-
pods but not larger animals), and gated (open to
both arthropods and rodents). After 4 months,
38%of theseedlingsinopencagesonSislandshad
been lost to herbivory, whereas losses at medium
and large landmass sites were lower at 18% and
14%, respectively (P=0.001). Survivorshi pin
skirted and gated cages did not differ signifi-
cantly and was lower than in impermeable cages
(P<0.005; Lopez and Terborgh 2007). These
results suggest that arthropod herbivory and not
seedling predation accounted for differences in
survivorship.Indirect effects and nutrient cyclingTree growth rates were over six times faster
on S islands supporting hyperdense howler
monkeys than on islands lacking howlers
(P<0.01). This seemingly counterintuitive asso-
ciation could result from an accelerated return
of plant available nutrients via monkey urine
and feces (recall that howler monkey biomass
was ca. 4000 kg km−^2 on some islands) as
opposed to slower pathways such as via leaf lit-
ter, throughfall, etc. (Feeley 2005b, Feeley and
Terborgh 2005). However, the enhancement of
tree growth is likely to be transitory. Intense