Tropical Forest Community Ecology

156 Stephen P. Hubbell

Janzen (1970) and Connell (1971) indepen-
dently suggested that an interaction between seed
dispersal and host-specific seed and seedling her-
bivores and predators could limit the local pop-
ulation density of tropical tree species, but their
focus was not on population regulation but on
explaining the high local tree species richness
of tropical forests by preventing any one species
from becoming monodominant. One can show
theoretically that Janzen–Connell effects do main-
tain more species locally at equilibrium (Chave
et al.2002, Hubbell and Lake 2003, Adler and
Muller-Landau 2005). However, Janzen–Connell
effects impose only a very weak dynamical con-
straint on the abundance of any given species
in species-rich communities. Imagine a “perfect”
Janzen–Connell effect that completely prevents
speciesifrom replacing itself in the same place in
the forest. The individual of speciesjthat replaces
a given tree of speciesican be any one ofS− 1
other species in the forest, each of which is not
constrained by the same enemy. If we turn this
argument around, speciesihas complete freedom
to replace any tree of theS−1 other species
in the forest, which occupy all sites not occu-
pied by speciesi. So long as speciesidoes not
approach monodominance, the dynamical con-
straint on the population growth of speciesi
is very weak. The more species-rich is the tree
community, the weaker the dynamical constraint
becomes on any given species.
Returning to the question of whether drift
or niche-assembly hypotheses better describe the
dynamics of the BCI tree community over the
past quarter century, the answer is once again
clear (Figure 9.6). There is an almost perfectly
linear decay in community similarity over time,
with no short-term evidence of a plateau of
community similarity – as measured by the
coefficient of determination of lagged species
composition. In contrast, the simulation of com-
munity dynamics under the stochastic equilib-
rium community under the hypothesis of niche
assembly clearly shows decelerating curvilinear-
ity and an approach to an asymptoticR^2 value. It
should be noted that ultimately the neutral model
decay curve also approaches an asymptote –
but much later and at a much lower positive
R^2 value than under niche assembly – set by

1.00

0.98

Mean

R

2

0.96

0.94

0.92

0.90

0.88

0 5 10 15

Lag (years)

Expected decay curve

under niche assembly

Expected decay under drift with q = 48, m = 0.09

Observed

20 25

Figure 9.6 Expected decay in community similarity,

as measured by the decay in the coefficient of

determination (R^2 ) of log species abundances at time

t+τauto-regressed on the log abundances of the same

species at timet. The black circles are the observedR^2

values, which fall along a straight line itself with anR^2

of 0.997. However, the straight line through the

observed values is not a regression, but is the prediction

from neutral theory derived from fitting the static

relative abundance data from the first census in 1982,

for whichθ=48 andm=0.09. The error bars are± 1

standard deviation of all possible 5-, 10-, 15-, and

20-year time lags between censuses. Lags involving

1982 were pooled with the nearest 5-year lag. In

contrast, the curve for the niche-assembly hypothesis,

assuming Gaussian stochastic variation carrying

capacities (lower cure), the niche-assembly hypothesis,

is not what is observed in the BCI tree community. Error

bars on the niche-assembly curve are±1 standard error

of the mean based on an ensemble of 100 simulations.

the immigration–extinction steady-state with the

metacommunity.

The drift prediction is actually much more

robust than simply demonstrating linear decay

in the coefficient of determination of commu-

nity similarity. We can predict quite precisely

the community dynamics from parameters mea-

sured only on the static relative abundance data

(Figure 9.6).The two key parameters from neutral

theory are the biodiversity numberθ(Fisher’sα)

and the immigration probability,m. These values

were estimated from the static first-census data for

1982 asθ=48 andm=0.09 (Table 9.1; Volkov

et al. 2005). The fit is essentially exact. This is a

powerful test of neutral theory in the case of the

BCI tropical forest, and it passed with flying colors.