32 M
.W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
Fig. 1. Measurement of the indices of structural complexity. a Transect showing profile of habitat profile
2
can be obtained from a real habitat or an experimental mimic. b Illustration of the calculation of VD, odh
n
and Chain at the 5-mm interval. VD is a measure of the variance in u ; VD 5 n 2 o a c
Y
n 2 1 , n is the
s h
s d
jd
s d
1
number of separate triangles along the transect. c Illustration of the calculation of D by the dividers method. The points are the apparent length of the transect measured by ‘dividers’ of increasingly greater intervals. The
equation for the line and the calculation of D are shown.
reduce the confusion that hinders our understanding of the effects of habitat structure on community structure.
2. Methods
2.1. Natural history All observations and experiments were done in Botany Bay, Australia on two rocky
intertidal shores, Bare Island and Sutherland Point Fig. 2. The study sites on these shores were on moderately sheltered, mid-shore sandstone rock benches. The rocks
appeared to be mostly bare with some encrusting algae; there was little or no foliose macroalgae. The obvious structural components were shallow pits often hemispherical
in the rock. Gastropods, mainly herbivores, were the most abundant, macroscopic animals in the study areas. Three species were common, the limpet Cellana tramoserica
and the snails Bembicium nanum and Austrocochlea porcata ex. constricta. A
. porcata
M .W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
33
Fig. 2. Map of Botany Bay showing rocky intertidal shores.
also occurs in nearby mangrove habitats, but the other species do not. There are detailed descriptions of these shores, their habitat structure, and the gastropods e.g. Underwood,
1975, 1976; Underwood and Chapman, 1996; Beck, 1998.
2.2. Habitat mimics and experimental design To test predictions from hypotheses 1–3, complexity and structural components were
2
manipulated with 0.12-m polyester resin plates Fig. 3. These plates were designed from computer models to have levels of complexity similar to those observed in rocky
intertidal habitats Beck, 1998 and to have components that were similar to those found in rocky intertidal and mangrove habitats Fig. 3. A gold pigment paste which was inert
when dried was added to the resin so that the color of the plates approximated the color of the natural rock surface rust-colored sandstone e.g. Walters and Wethey, 1996. The
plates were fastened on to the shore with stainless steel screws that were countersunk into the plates and screwed into wall plugs in the rock.
Complexity was measured with a variety of indices Fig. 3. The calculation of D, VD,
2
odh , and Chain is explained in Fig. 1 and Beck 1998. It is necessary to choose a clear scale for measurements of complexity McCoy and Bell, 1991. In prior work it was
shown that the gastropods at these sites were most affected by complexity measured at 5-mm intervals as compared to greater intervals Beck, 1998. Therefore, 5 mm was the
finest scale of measurement used in the development of these treatments. Only gastropods larger than 5 mm in size were included in analyses. This size cutoff also
helped to focus analyses on gastropods that moved onto treatments not larvae that settled on them. There were few gastropod or barnacle settlers on the plates during the
experiments.
These plates were used to test one assumption about the suitability of the plates as
34 M
.W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
Fig. 3. Experimental design for manipulation of complexity and structural components. Values underneath treatments indicate their complexity as computed by different indices. The low complexity treatments are
coded so that the first letter indicates diameter 4 or 6 cm and the second indicates depth of the pits 0.75, 1.0, or 1.5 cm; WM5wide, medium; NS5narrow, shallow; ND5narrow, deep; Pneum5pneumatophores. Pits on
the High complexity treatment were 4 and 6 cm in diameter and 1.7 cm deep. Pneumatophores were 3 cm tall and 1.0 cm in base diameter. These dimensions were similar to the dimensions of natural structural
components in these habitats.
habitat mimics and hypotheses 1–3 from the Introduction. The response variables were
2 2
the density no. of individuals 0.12 m and richness no. of species 0.12 m of gastropods on each treatment.
2.2.1. Assumption 1: The plates were reasonable mimics of the habitat If this assumption is correct, it is predicted that the density and richness of gastropods
on the high and low complexity treatments would be similar to the density and richness of gastropods in undisturbed reference quadrats 5 control treatment on the surround-
ing rocky intertidal platform. The high and low complexity plates were explicitly designed to have complexities Beck, 1998 and components similar to those measured
previously on the rocky intertidal platforms at these sites. The reference quadrats had the same dimensions as the plates 30 3 40 cm and their locations were haphazardly
determined.
2.2.2. Hypothesis 1: Habitat complexity positively affects the density and richness of gastropods
If this hypothesis is correct, the density and richness of gastropods is predicted to be greater in the high complexity treatment than in the group of three low complexity
treatments with pits.
M .W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
35
2.2.3. Hypothesis 2: D represents elements of complexity that affect the density and
2
richness of gastropods better than other indices , i.e. VD,
odh , surface area, number of pits
The three low complexity treatments with pits were explicitly designed to separate among these indices. If D best represents elements of complexity that affect gastropods,
as predicted by Beck 1998, their density and richness on the treatments should be NS , ND , WM Table 1, Fig. 3. The other indices make different predictions about
the rank order of density and richness on treatments Table 1, Fig. 3. D and Chain measure complexity in similar ways Beck, 1998, and it was not possible to construct
manipulations from which contrasting predictions about their effects on the density and richness of gastropods could be derived.
2.2.4. Hypothesis 3: Specific components have effects on the density and richness of gastropods that are independent of complexity
This hypothesis is tested by manipulating only structural components and holding complexity constant. Mimics of pits and pneumatophores of the mangrove Avicennia
marina were used to manipulate structural components. Pneumatophores were chosen, because i they are a definitively different component from pits, ii gastropods occur
on pneumatophores on mangrove shores around Sydney, and iii in a companion study, manipulations of the complexity of pneumatophores affected the density and diversity of
gastropods in mangrove habitats Beck, unpublished data. If this hypothesis is correct, the density and richness of gastropods is predicted to be greater on the three low
complexity treatments with pits than on the treatment with pneumatophores. It was not possible to hold all indices of complexity constant on these treatments Fig. 3, and an
emphasis was placed on holding SA and D constant based in part on prior results Beck, 1998.
Experiments with these treatments were repeated as many as five different times during 1996 and 1997 at two different sites within each of the two shores. There were
two replicates of each treatment at each site used in experiments. Not all treatments, sites, and shores were used in each experiment Appendix 1. Experiment 3 included all
treatments at all sites and shores, but during the course of the experiment vandals destroyed all of the plates at Sutherland Point site 2. These plates were not replaced, and
a second site was not used in experiments 3, 4, or 5 at Sutherland Point. Before each
Table 1
a
Predicted responses of gastropods to low complexity treatments based on different indices of complexity Index
Predicted density and richness D
NS,ND,WM VD
ND,NS,WM
2
odh NS,ND5WM
Chain NS,ND,WM
SA NS5ND,WM
No. of pits ND,WM,NS
a
The specific values of the indices on these treatments are in Fig. 3.
36 M
.W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
experiment, plates were scrubbed clean and randomly allocated to positions at the sites. Each experiment lasted approximately two months with variation in duration because
weather and tides limited accessibility. The five experiments were concluded in August, December 1996, February, April, and
June 1997. The experiments examine the reproducibility and temporal consistency of these results at different random times; they were not intended to test seasonal
hypotheses per se. The experiments do, however, cover periods with different prevailing climatic conditions.
2.3. Data analysis The total density and richness of gastropods were analyzed for each experiment by
ANOVA. Treatment was a fixed factor and site was a random factor. In experiment 2, site was nested within shore, which was a random factor. The factor, shore, could not be
used in ANOVAs for experiments 3–5, because the plates at the second site at Sutherland Point were destroyed. Densities were transformed log x to meet conditions
e
of homoscedasticity. In experiment 4, one replicate plate was lost, and in experiment 5, four plates were lost in storms Appendix 1. To balance the ANOVAs for these missing
replicates, the data were replaced with the value from the other replicate at that time and place. The residual df were reduced by 1 and 4, respectively. Mean square values were
obtained from SAS Release 6.04, SAS Institute Inc., Cary, NC, USA.
When the effect of treatment was significant, a priori contrasts were done to test the predictions of the assumption and hypotheses 1–3 above. To test the predictions from
hypothesis 2 Table 1, it was necessary to do all the pairwise comparisons among the three low complexity treatments. These comparisons were non-orthogonal. To hold the
type I error constant at a 5 0.05 for this group of comparisons each comparison was tested at a 9 5 0.017 following Dunn–Sidak’s procedure Underwood, 1997. When there
were significant treatment 3 site interactions, contrasts were done within each site with Student–Newman–Keuls SNK tests Underwood, 1997.
2.4. Combining data from experiments. These separate experiments provided independent tests of the predictions identified
above, and it was possible with rank-order statistics to combine the data across the different experiments. Data were combined across experiments, because in some cases
the results from individual experiments did not indicate significant differences among treatments, but there appeared to be clear patterns at most times and places. It also was
possible to summarize the results briefly and clearly by combining data across experiments. In addition, it was possible to examine the effects of the treatments on the
densities of the three most abundant species across the experiments. ANOVAs were not done on the densities of the individual species, because these data did not meet the
assumptions of the test.
Two non-parametric tests were used to combine data; one proposed originally by Anderson Anderson, 1959; Winer et al., 1991, which will hereafter be referred to as
2
Anderson’s Q and a binomial test. The basis of both these tests is that responses to
M .W. Beck J. Exp. Mar. Biol. Ecol. 249 2000 29 –49
37
treatments are ranked at the different times and places trials for each contrast or comparison, and the likelihood that the rankings are randomly distributed among
treatments is assessed. In these tests, each site in each experiment was a separate trial i.e. each row in Appendix 1 represents a separate trial, and the responses to the
treatments were ranked within each trial for each of the a priori contrasts identified in assumption 1 and hypotheses 1–3. There was thus a maximum of 14 possible trials at
separate places and times Appendix 1.
2
Anderson’s Q was used to examine distributions of ranks in the contrasts one vs. three treatments when there were no tied ranks. In cases with tied ranks and for
comparisons e.g. high vs. control, binomial tests were used to combine results across experiments. If there was no difference in density or richness between treatments, each
treatment should be ranked first on average 50 of the time i.e. binomial p 5 q 5 0.5. When the binomial test was used for contrasts that involved the three low complexity
treatments with pits i.e. low vs. control, high vs. low, pits vs. pneumatophores, density and richness were averaged among the three treatments to condense the contrast to a
comparison of two groups. In trials with tied values, a tie was counted against the predictions i.e. in support of the null prediction if there was just one tie. When there
were two tied values, one value was counted in support of predictions and one against. In the analyses of the a priori contrasts for the individual species, a few trials were
dropped from consideration, because no individuals of that species were observed on any treatment in the trial. The number of omitted trials was noted on every occasion when
this procedure was done.
3. Results