Fróðskaparrit - 31.12.2000, Blaðsíða 104

108
DOMINANT SPECIES ABUNDANCE RELATED TO ENVIRONMENTAL
FACTORS ON ROCKY SHORES IN THE FAROEISLANDS
FEV values of Bruntse et al. (1999b), ex-
cept that the scale is inverted so that high
values signify high exposure. The shore as-
pect variable is coded to reflect expected
differences in amounts of sunlight received
(Table 2).
Numerical methods
Canonical Correspondence Analysis
(CCA) was used to examine the relation-
ships between species and environmental
factors. CCA is an ordination technique
that maximises the dispersion of species
centroids (the weighted averages of the dif-
ferent species) along axes that are con-
strained to be linear combinations of the
environmental variables (ter Braak, 1986;
ter Braak and Verdonschot, 1995). The
method is based on a unimodal response
model (e.g. ter Braak, 1995). If the com-
munity variation is within a narrow range,
linear ordination methods (Principal Com-
ponent Analysis and Redundancy Analy-
sis) are appropriate because most species
are behaving monotonically over the ob-
served range (ter Braak and Prentice,
1988). The gradient length of the first axis
in Detrended Correspondence Analysis
(DCA), measured in standard deviation
units of turnover (SD), indicates which
method to use. For gradients less than about
1.5-3 SD, the approximations involved in
weighted averaging (used in CCA, DCA
and Correspondence Analysis, CA) be-
come worse (ter Braak and Prentice, 1988).
The first DCA axis for the present data set
was 2.9 SD, giving no clear indication of
which model to apply. The weighted-aver-
aging methods were chosen since with this
data the main DCA axis explained a higher
percentage of the species data than the
main axis in Principal Component Analy-
sis.
The eigenvalue of an axis in CCA is a
measure of the amount of variation ex-
plained by it. The total variance is given by
the sum of all unconstrained eigenvalues in
a Correspondence Analysis (CA). The im-
portance of an environmental variable in
the ordination may be expressed by the
amount of variance (referred to as inertia)
attributable to it. If the environmental vari-
ables are inter-correlated, the correlation of
each variable with the major canonical axes
may better indicate its significance (ter
Braak, 1986; ter Braak and Verdonschot,
1995). This can be shown in an ordination
diagram in which each environmental vari-
able is shown as a vector from the origin
(centre) to a point (x, y), in which x and y
approximate the correlation between the
variable and two given canonical axes. The
species centroids can be shown in the same
diagram, and each centroid can be project-
ed to either canonical axis or any environ-
mental variable to find the species’ weight-
ed average score on the axis or variable (ter
Braak and Verdonschot, 1995). In the ana-
lyses, the environmental variables of im-
portance were selected by a ”forward selec-
tion” procedure. The method ranks the
variables in importance and selects them
one by one starting with the one that would
add the most inertia if included. For each
step, the significance of the new variable
was tested using the Monte Carlo Permuta-
tion Test (999 unrestricted permutations),
and the variable was included if significant