One of the major questions in ecology is, what controls the structure
of communities? We used projection matrix models to examine community
dynamics and patterns of succession. The inputs of the model are trans
ition probabilities of species replacements that were measured repeate
dly during a long-term (1962-1989) study of diverse coral assemblages
on Heron Island, Great Barrier Reef. Transitions varied strikingly amo
ng species and sites, reflecting differences in recruitment, growth, l
ongevity (persistence), and the rate of replacement of one species by
another. Species that had a poor ability to persist (e.g., algae and P
ocilloporid corals) were generally good colonists. The observed number
of transitions expressed as a proportion of the maximum number possib
le provides an index of the complexity of interactions in an assemblag
e, analogous to the concept of connectance in food-web analysis. Trans
itions occurred to and from nearly every species group, indicating tha
t there was no competitive dominant in this system. We use the models
in simulations to track transitory changes in species abundance and co
mmunity composition following a major disturbance (e.g., due to a cycl
one or outbreak of crown-of-thorns starfish). Some species showed a ra
pid initial increase followed by a decline to lower equilibrium levels
, while others increased smoothly to a generally higher equilibrial ab
undance. The length of time required to reach a climax assemblage usin
g the same matrix recurrently (approximate to 20 yr) is far greater th
an the observed interval between major disturbances, supporting nonequ
ilibrium theories of coral reef communities. Climax assemblages were h
ighly diverse and varied in composition from site to site. The ''inter
mediate disturbance hypothesis'' does not fully predict successional c
hanges in these shallow-water coral assemblages since diversity remain
ed very high at equilibrium (i.e., long after a major disturbance), Co
mpetitively inferior species were not eliminated because routine morta
lity ensured that some space always remained available for colonizatio
n. We also present a novel method for quantifying the relative importa
nce of each species interaction to community composition and the rate
of succession, based on a sensitivity analysis of the transition matri
x. The analysis shows that the importance of a species to the dynamics
of a community may be unrelated to its abundance at equilibrium, with
some rare species groups having a greater impact than more common one
s. Sensitivity analysis of this type will provide a powerful means of
identifying ''keystone'' species in complex assemblages where experime
ntal manipulation of each species is impossible.