Tropical Forest Community Ecology

154 Stephen P. Hubbell

Table 9.2 Changes in abundance in Barro Colorado Island tree species in the 50 ha plot over a

23-year interval from 1982 to 2005.

% absolute change

in abundance,

1982–2005

No. of species Mean 1982 species

abundance

Standard deviation

in 1982 species

abundance

0–25 128 1379.6 4380.7

25–50 66 568.0 1038.7

50–75 36 306. 4622.8

75–100 2 4317.5 65 4.1

> 100 31 83.1 154.9

Notes: Only the 285 species with two or more individuals in 1982 are included. Percentage change was calcu-

lated as the absolute value of the difference in abundance of a species in 2005 and 1982 times 100 divided by

the abundance of the species in 1982. The mean percent change±1 standard deviation over all species was

47.6±67.0%.

sufficiently long to perform such a test. One of
these is the plot on Barro Colorado Island, for
which we have a quarter-century record of for-
est change and turnover (Hubbell 2008c). The
BCI forest has exhibited remarkable dynamism
over the last 25 years (Table 9.2). Of the 285
species sufficiently abundant to test, nearly a third
(31.9%) changed by more than 50% in total abun-
dance, and these changes were not limited to rare
species.These changes are not purely successional
because two-thirds of the species (63.9%) do not
exhibit monotonic directional changes in abun-
dance. In and of itself, this level of dynamism
would be challenging to most equilibrium the-
ories of community organization. However, we
need to test the possibility that these changes are
simply Gaussian stochastic fluctuations centrally
tending around fixed carrying capacities in an
equilibrium, niche-assembled community. Alter-
natively, these fluctuations may be more accu-
rately described by drift, with no central tendency,
the prediction of neutral theory.
A simple measure of community change over
time is the decay in the coefficient of determi-
nation,R^2 , of community composition. In order
to normalize changes on a per capita basis, one
should evaluate changes in abundance on a log
scale. At a time lag of zero, no change can yet
have occurred, and the auto-regression of species
abundances on themselves therefore yields an
R^2 of unity. But as the time between censuses

increases, we expect a decay in the value ofR^2 of

the regression of log species abundances at time

t+τon the log abundances of the same species

at previous timet. We can test the predictions

of niche assembly versus neutrality for commu-

nity dynamics because these theories make very

different predictions of the expected patterns of

decay inR^2 over time. The prediction from neu-

tral theory, when the metacommunity is very

species-rich, is nearly perfectly linear decay in

similarity (as measured byR^2 of the time-lagged

regression of log abundances). In contrast, under

Gaussian stochastic fluctuations around a sta-

ble equilibrium of a niche-assembled community,

one expects a relatively fast, curvilinear decay

inR^2 , to an asymptoticR^2 value, reflecting the

underlying community stability and the tendency

of species to approach their niche-determined

carrying capacities.

To test the niche-assembly predictions, we

randomly sampled the distribution of observed

intrinsic rates of increase of the BCI species

over the past quarter century, which are approxi-

mately normally distributed with a mean of zero

(Figure 9.4). We then applied the randomly sam-

pled intrinsic rates, normalized to 5-year census

intervals, to each species and computed the decay

inR^2 of log species abundances over 25 years and

six censuses. We repeated this procedure for an

ensemble of 1000 stochastic runs, and averaged

the results.