van Etten & Fox: Vegetation classification - Hamerlsey Ranges 
woodlands with bunch grasses (community 6A). These 
two coolabah-mulga sites had the lowest affinities to their 
group of all sites and indeed formed a separate group 
when the ALOC procedure was forced to produce 17 
groups. 
Classifications using an initial matrix of transformed 
species cover values rather than lVs also resulted in very 
similar groupings of sites to that described above, and, 
for most communities, their composition in terms of sites 
were identical. The two major differences of the 
dendrogram produced using cover values were: 1) 
community 4B contained several more sites dominated 
by T. wiseana including several which were not on 
volcanic plains but on hill tops and upper slopes of 
Brockman ironstone formation where several of the 
characteristic shrubs of this formation are often absent; 
and 2) community 7 A was more clearly divided into two 
groups, one with the ground-layer dominated by T. 
pangens and the other by T. triandra and other tussock 
grasses. This synchrony between the two classification 
approaches is no doubt derived from the strong 
correlation between transformed IVs and cover values 
(r=0.87; n=190). 
Analysis of similarity (ANOSIM) test, performed with 
1000 random permutations, demonstrated that the 16 
groups determined by the two classification procedures 
were always more dissimilar to each other than 16 
groups randomly determined. This implies that groups 
determined are significantly different to one another in 
terms of species composition (p<0.001). Although this is 
expected, the ANOSIM, and the fact that two completely 
different classification algorithms (one hierarchical and 
the other not) produced very similar groupings, support 
the idea that "real" groupings were detected and suggest 
that correct decisions were made with respect to 
classification methodology (Belbin 1994). 
Community comparisons 
A comparison of structural, floristic and growth-form 
features are displayed in Table 3 for the plant 
communities. There are several significant structural 
differences between the various plant communities and 
vegetation types. Mean total cover varies from 22% on 
mountain tops to 42-50% on drainage lines and flats. 
Mulga woodland on flats (community 6A) and E. 
xerothermica woodland within drainage lines (7A) have 
significantly higher cover than hummock grassland 
communities (Table 3). Tree cover is, not surprisingly, 
significantly higher in woodland communities compared 
to communities of hummock grassland on slopes and 
pediments (typically 15-24% cover compared to between 
4-6% cover of trees). Tree cover on volcanics (community 
4B) is significantly lower on average than many other 
communities, including some hummock grassland ones. 
Grass cover is relatively high (20-30%), except in the 
following communities which are significantly lower 
than most others: mulga woodland on scree slopes (4C), 
snakewood woodlands (3B), hummock grassland on 
mountain tops (4A) and mulga - T. melvillei - tussock 
grass (6B). 
Average (perennial) species richness and diversity are 
significantly higher within mulga woodland on flats 
(community 6A) and creekline E. camaldulensis - E. victrix 
woodlands (7B) than in most other communities (Table 
3). In addition, far more annual/ephemeral species were 
sampled in plots of these two communities than in 
hummock grassland communities. Hummock grassland 
on pediments and bajada slopes (communities 2A, 2B & 
5) are generally less rich and diverse than other 
communities. Community 8 (E. victrix woodland of 
alluvial basins) had the lowest perennial richness and 
diversity of all communities (Table 3); median values 
were significantly lower than most other communities. 
Diversity values, although heavily influenced by species 
richness, also depend on the evenness of species 
distribution at sites. Values of evenness are relatively 
similar for all communities (i.e. around 0.6-0.7), except 
for communities 6A and 7B which are significantly 
higher (i.e. more even distribution of species) and 2B 
which is significantly lower than most other communities 
(Table 3). The richness of uncommon species (those 
found at less than 3 sites) is significantly higher (on 
average) within mulga woodland on flats (community 
6A), along creeklines (community 7B) and on mountain 
tops (community 4A) compared to many other 
communities (Table 3). Whilst actual numbers would 
depend to a certain degree on the intensity of sampling 
in each community, these three communities have far 
more uncommon species than other similarly sampled 
communities. The uncommon species are mainly herbs 
and subshrubs in the first two communities, whilst they 
are mostly low shrubs on the mountain tops. Species 
positively identified as being rare and/or poorly known 
(as listed by the Western Australian Department of 
Conservation and Land Management) were few and 
mainly found on mountain tops. 
Communities on calcareous soils (3A & 3B), T. wiseana 
hummock grassland on volcanics (4B) and E. xerothermica 
drainage line community (7A) are much less 
homogeneous in terms of species composition than other 
communities (Table 3). Each of these communities has at 
least one site that is distinct in some way to all the others. 
For instance, community 4B includes one site containing 
no trees on calcareous soils which seems to be placed in 
this group because of shared dominance by T. wiseana. 
Community 3B includes two quite dissimilar sites 
dominated by A. victoriae and other Acacia shrubs, rather 
than A. xiphophylla. They are clumped with other 3B sites 
because of understorey similarities. Quite possibly, these 
four heterogeneous communities will be split into a 
number of quite distinctive sub-communities, given the 
patterns shown in the classification, but sampling has 
been insufficient to confirm this. Communities which 
were relatively homogeneous in terms of species 
composition are 8,1 A, IB and 2B (Table 3). 
Ordinations of sites 
The three-dimensional solution provided a 
considerable improvement in stress over a two 
dimensional solution (0.18 c.f. 0.26) and were hence used 
despite the increased difficulty in displaying patterns 
(Fig 3). Kruskal (1964) suggests that stress values less 
than 0.2 should be aimed for, whilst values less than 0.1 
provide a "satisfactory" goodness of fit. These however 
are general guidelines at best as stress values are 
influenced by, amongst other things, number of sites and 
heterogeneity of communities. Ordinations performed 
using cover values rather than IVs in the initial site by 
site matrix produced a very similar pattern of 
73 
