112 Jess K. Zimmermanetal.
was a key factor in the maintenance of diversity
of trees and shrubs in tropical forests (e.g., Grubb
1977, Denslow 1987).
Habitat specialization beyond gap dependence
could provide another explanation of high species
diversity. In the relatively flat plot at BCI, which
occupies the top of a hill, the scope of habitat spe-
cialization appeared limited (Hubbell and Foster
1986, Harmset al. 2000) although a handful of
species occurred only in a swampy area and along
slopes.Valenciaet al.(2004b)recentlyinvestigated
topographic habitat specialization in the more
topographically varied Yasuní plot in Ecuador.
There was some evidence of niche partitioning
among valley, mid-slope, and upper ridge areas.
While about 25% of species had large abundance
differences among topographic positions, another
25% were complete generalists.Thus, partitioning
among topographic niches at Yasuní provided no
explanationof theco-occurrenceof generalists,or
the hundreds of species sharing the ridge-tops. In
the case of theYasuní FDP, Valenciaet al.(2004b)
concluded that this type of habitat specialization
makes only a minor contribution to local species
diversity. Studies of habitat specialization in more
FDPs will be needed before we fully understand the
precise role of habitat specialization in explaining
the species diversity of tropical forests.
Overall, the CTFS network of large FDPs has
done a great deal to distinguish among three
prominent explanations for the diversity of trop-
ical forests. Analyses of data from the CTFS
network of large FDPs make clear that there is
some rare species advantage common to trop-
ical forests and also some evidence for habitat
preference that promotes species diversity. The
evidence from the network of plots does not
support a strong role for gap specialization in
explaining patterns of tree and shrub species
diversity, and more information is needed on
other potential habitat niches that may influence
diversity. The results obtained so far in the FDPs
suggest that on balance species diversity in trop-
ical forests is not consistent with the null model
of species interactions provided by the neutral
theory of Hubbell (2001). In addition, the phe-
nomenological data collected in the FDPs must be
substantiated by direct experimental tests of the
mechanisms involved.
LIMITATIONS
The large-plot approach has some limitations
with regard to testing ecological theory and here
we briefly touch on these. Some ecologists object
to the use of one large plot because individ-
ual contiguous plots within one large forested
area are “pseudoreplicated” because there are no
randomly and independently selected replicates
(Hurlbert 1984, Scheiner 2001).Thus, the results
may not be generalized to any forested areaaway
from the plot. The only way to correct this would
be to have replicated large plots, perhaps of a
smaller size, randomly located throughout the
forest. This presents obvious logistical difficulties
and places limits on the types of questions that
could be addressed (e.g., seed dispersal distance,
or neighborhood effects) in several smaller plots.
Small plots suffer from the edge effect, insufficient
numbers of complete tree neighborhoods, and an
absence of data from local tree neighborhoods
that extend outside of the plot edges. Sampling
smaller plotsawayfrom an FDP may be effectively
used to place the FDP in a larger, regional con-
text of forest variability (e.g., Conditet al. 2002).
Statistical techniques that take into account spa-
tialautocorrelation(i.e.,thelackof independence)
of trees or subplots can also be employed to
address this problem (Robertson 1987, Rossiet al.
1992, Hubbellet al. 2001). Thus, while not insur-
mountable, it is a real problem that each FDP is a
single sample, albeit a very large one.The strength
in the approach, however, lies in the network of
plots throughout the world. While not true repli-
cates in a statistical sense, robust results from a
variety of forest types, such as those provided by
Willset al. (2006) and Conditet al. (2006), support
the overall approach.
Another limitation of using FDPs to test
hypotheses explaining the high diversity of their
forest communities is that the results are phe-
nomenological (Hubbell 2004). They thus typi-
cally do not provide evidence that a particular
mechanism is operating (e.g., the Janzen–Connell
hypothesis), and the observed patterns could be
the result of other factors not considered. For
example, Willset al. (2006) cite three factors that
may explain increasing plot diversity with time.
Nonetheless, the consistent lack of evidence for