Spatial Variation in Tree Species Composition 21the mid-1990s. Terborghet al.’s (1996) test
of Hubbell’s (1979) non-equilibrium hypothesis
for tropical forests played an important role in
this progress. Unlike previous works, Terborgh
et al. (1996) addressed the relevance of the non-
equilibrium hypothesis at the landscape scale, not
just locally as had previously been done. They
identified all trees greater than 10 cm diameter in
several plots alon gthe río Manu, southeast Peru,
and showed that a few species were both abun-
dant and widespread (i.e., “oligarchic species,”
sensuPitmanet al. 1999). Thus Hubbell’s non-
equilibrium theory, even if valid at the local scale,
cannot be true at a larger scale. Hubbell (1997)
then extended his local neutral model to a “two-
scale” model where the local community interacts
with a regional species pool in which diversity
is maintained by evolutionary processes (see also
Hubbell 2001, 2003, Ricklefs 2003, Chave 2004).
Thus, by allowin gfor predictions at a lar ger scale,
the neutral theory was significantly improved.
Hubbell’s (1997) theory viewed each local com-
munity as one sample of a regional species pool,
hence allowin gfor re gionally abundant species
to be abundant in local samples. Hubbell (2001)
further expanded this idea, showin gthat the
mechanism of seed dispersal limitation can be
included in a neutral theory and may contribute
to explain the decrease of species similarity with
distance across a landscape. This aspect was for-
malized in greater mathematical detail by Chave
and Leigh (2002).
Conditet al. (2002) brought theory together
with empirical data usin gpermanent tree plot
censuses previously assembled by three indepen-
dent research groups (Terborghet al. 1996 in
Peru, Pitmanet al. 2001 in Ecuador, Pykeet al.
2001 in Panama). More precisely, they tested one
of the predictions of Chave and Leigh’s (2002)
model described above, that the similarity func-
tionF(r)should decrease logarithmically with
increasin gdistance (Equation (2.6)). Usin gsim-
ple regression approaches, Conditet al. found that
a logarithmic function indeed provided a correct
fit forF(r)across several decades forr, from 0.1
to 100 km (Figure 2.4). They were also able to
estimate the value of the model’s parameters,
and found values of the dispersal parameterσ
of about 50 m, consistent with known estimates
for the dispersal of tropical trees. The value for
the speciation rateνwas also estimated, but with
rather poor accuracy (on the order of 10−^14
to 10−^8 ). This was the first explicit test of the
influence of distance in the shapin gof tree com-
munitiesatthelandscapescale(seealsoHardyand
Sonké 2004 for a similar analysis in the tropical
forests of Cameroon).
Interestingly, the Chave and Leigh model failed
to fit the observations at both short and large
distances. The large-distance discrepancy can be
easily explained by the fact that Equation 2.6
is only an approximation of the exact formula
predicted in Chave and Leigh (2002). The short-
distance discrepancy is more troublesome: in
most cases the similarity at short distance is far
greater than that predicted by the neutral model.
At least two, non-mutually exclusive, explana-
tions can be provided for this discrepancy. First,
the neutral model analyzed by Chave and Leigh
(2002)assumesaGaussiandispersalkernelwhich
is convenient but may not provide the best fit of
real dispersal kernel at short distance. Second,
other biological processes may be acting at this
local scale, for example density-dependent factors
(T. Zillioet al. unpublished results, and S.P. Hubbell
personal communication) or habitat specializa-
tion. It is premature to decide which explanation
is the more likely, and more research on this topic
is necessary.
Another surprisin gfeature of the Condit
et al. (2002) study is the strikin gdifference in
the similarity functions in the large plot data
from Panama and western Amazon (Peru and
Ecuador). The Panamanian similarity function
decreased steadily with increasin glo garithmic
distance. In contrast, the western Amazon simi-
larity function decreased rapidly for distances less
than 100 m, and much more slowly at larger dis-
tances. One interpretation for this pattern is that
the histories of the two forests are different: tree
species may have had more time to disperse in the
western Amazon than in Panama (Conditet al.
2002).Partitioning the effects of dispersal
and environmentDuivenvoordenet al. (2002) pointed out that
the differences between Panama and the western