van Etten & Fox: Vegetation classification - Hamerlsey Ranges 
found throughout the geological profile, particularly in 
the softer, dolomitic formations. The surface geology is 
therefore somewhat complex with the main geological 
formations of the study area surface being the Brockman 
Iron Formation, the older Marra Mamba Iron Formation, 
the Jerrinah Formation (very complex and dominated by 
dolomite and shale), the Weeli Wolli Formation (also 
dolomite and shale), Archaean granite, volcanics. Tertiary 
consolidated colluvium, the Oakover Formation (Tertiary 
calcareous gravels and limestone). Quaternary colluvium, 
and Quaternary alluvium (Thorne & Tyler 1997). 
No long-term rainfall records are available for the 
study area. However, interpolation of long-term records 
from surrounding stations suggests rainfall for the study 
area is around 350 mm per annum on average (Beard 
1975; van Etten 2000) and decreases in a southerly 
direction. This average figure however belies the very 
large temporal variability in rainfall across the region; 
the variability index of annual rainfall is between 1 and 
1.2, which is classified as moderate to high (Bureau of 
Meteorology 1989). This variability mainly stems from 
the erratic nature of cyclonic rains during summer and 
autumn. In some years, cyclones cross the coast and 
bring large amounts of rain to the arid interior, whereas 
in other years no effective cyclonic rains may occur 
(Beard 1975). Winter rainfall averages around 40-50 mm, 
increases with distance south and is much more regular 
than summer rain. Temperatures and evaporation rates 
are high in summer with the mean January maximum 
close to 40°C. Winter maximum temperatures average 
around 24°C, with average minimum around 11°C. Mild 
frosts have regularly been observed during winter (Beard 
1975), particularly in valleys and hollows. 
Study sites & field measurements 
Truly random selection of study sites was not possible 
given the lack of roads and tracks, the restrictions on off¬ 
road vehicle use and the ruggedness of the terrain. 
Alternatively, available roads and vehicle tracks were 
used to sample as much variation in landform type, 
geological substrate and altitude as possible. Altogether 
139 sites were chosen for study. All major landform types 
and geological substrates were sampled several times 
throughout the study area. Areas immediately adjacent 
to roads were avoided because of the potential for edge 
effects, as were recently burnt areas given they have a 
suite of pioneer species. Sampling occurred over seven 
separate periods: June 1991, August 1991, December 1991, 
July 1993, December 1993, December 1994 and July 1995. 
The amount of rainfall before sampling, which influences 
annual and ephemeral species composition, varied 
considerably. 
At each study site, a 20 m x 25 m plot was marked out 
on the ground and then divided into twenty 5 m x 5 m 
quadrats. This size plot (0.05 ha) was used as it has been 
recommended for studies of arid woodland communities 
(Fox 1981; Kent & Coker 1992) and supported by 
examination of species-area curves which showed they 
regularly capture the majority of species present at sites. 
In each 25m 2 quadrat, the abundance (number of 
individuals) and cover of each perennial species was 
estimated. Perennial species in this study included short¬ 
lived perennials (sometimes referred to as semi¬ 
perennials) which can survive longer than a year given 
favourable conditions. Most species of semi-perennials 
were present at all sampling times over the course of the 
study, although their abundance and the amount of 
green tissue varied somewhat. True annual species (i.e. 
those which consistently complete their life-cycles within 
a year) were not included in the analyses given their 
temporal variability. Species were identified using 
relevant keys and the W.A. Herbarium (Perth) collection 
with the authority for plant names and nomenclature 
being Paczkowska & Chapman (2000). Voucher 
specimens of all species collected are lodged in the Edith 
Cowan University Herbarium (Joondalup Campus). 
Data analysis 
An importance value (IV) was calculated for each 
perennial species at each plot by summing the relative 
cover, density and frequency of plants and multiplying 
by 100 (Curtis 1959). The sum of IVs for all species in a 
plot is 300. The IV of each species in the plot is therefore 
relative to the other species in the plot and is not an 
absolute measure. The main advantage of the importance 
value over other measures of cover/abundance is that it 
enables fair comparison of different sized plants. For 
instance a tree species (typically high cover, but low 
abundance) can have similar IVs to grasses/herbs/ 
subshrubs (low cover, high abundance), but would differ 
considerably if either cover or abundance measures only 
were used. IVs were log transformed using the formula 
x' = log e (x + 1) to downweight dominant species and to 
reduce the skewness of the data set - this effectively 
reduced the data to between 0.1 to 5. A site by species 
matrix was then constructed using transformed IVs and 
converted to a site X site dissimilarity matrix using the 
Bray-Curtis association measure (Bray & Curtis 1957) as 
recommended by Faith et al. (1987). Classification of sites 
was performed on the dissimilarity matrix using two 
different clustering strategies: ALOC (Belbin 1987) and 
flexible-UPGMA (Belbin et al. 1992). For the flexible 
version of UPGMA, b values of -0.1, as recommended by 
Belbin & McDonald (1993), were used. Ordination of the 
site dissimilarity matrix was performed using a hybrid 
multidimensional scaling method (SSH) with the "ratio- 
ordinal cut value" set at around 0.8 as suggested by 
locating a trough in the histogram of association values 
(Belbin 1991, 1994). In addition to ordination of all 139 
sites, 35 sites located along a particular ridge-top to 
valley floor toposequence near Marandoo (Fig 1) were 
ordinated and a differential profile constructed by 
plotting the one-dimensional SSH scores against distance 
along topo-sequence (Hobbs 1986). Principal axis 
correlation (PCC) was used to calculate how well each 
species was linearly correlated to the ordination space 
and the direction of this correlation (Belbin 1994). To 
protect against spurious results using IV's, the same 
classification and ordination procedures were also 
undertaken using cover values (square root transformed) 
and compared to results using IV's. 
Mean cover of all species and of various life-forms 
were compared for various vegetation units using one¬ 
way ANOVA, with Tukeys post-hoc test used to 
distinguish which pairs of means were significantly 
different. Means were first checked for equal variances 
and all cover values were arcsine transformed before 
analysis (Sokal & Rohlf 1981). Mean community species 
richness (of all species, as well as that of various life- 
65 
