Free Download - HCCREMS

3.4.5 Classification of the survey data
CHaPTeR 3 Method of data analysis and collection
Classification analyses for this study were primarily carried out within the DOS version of
the PATN software package (Belbin 1994). Additional analyses, such as fidelity analysis
(Keith & Bedward 1999), were carried out using queries in Microsoft Office Access (Microsoft
Corporation 2003). How the analysis dataset that was used in the classification process was
derived is described above in section 3.4.4.
For all analyses other than the preliminary exploratory analyses, the Bray-Curtis measure of
association was used due to its appropriateness when analysing cover / abundance scores (see
Section 3.3.1 above).
An initial agglomerative hierarchical classification of the data into groups was carried out
using the FUSE module in PATN (Belbin 2005). A Flexible UPGMA strategy was applied to this
classification with a beta left at the software default of -0.1. This approach was used to produce
150 and 300 groups as it was felt that the final number of communities would be between these
values; this allowed for an understanding of the broader patterns and relationships between
the finer groups to assist in the merging of these groups.
Initial analyses of the outputs of this first iteration of the classification suggested further issues
with the dataset which had not previously been detected. The dataset was further refined,
with some sites being removed due to suspected biases or errors, and cover scores for some
data being re-standardised. In addition to these refinements, data from all groups which had
been defined for sites with less than 10 species were removed before the second iteration to
remove any possible effects these sites may have had on the overall classification. These groups
were reintroduced to the classification once the remaining data had been interpreted. A second
iteration of the hierarchical classification of 150 and 300 groups was then carried out using the
FUSE module in PATN.
In parallel to this approach, a non-hierarchical classification was also carried out using the ALOC
module in PATN (Belbin 1995) in order to compare the outputs from these two approaches. Nonhierarchical
classification has been suggested as more appropriate for use in cases where a very
large dataset is to be analysed (Belbin 1995); however, in this case the groups produced were
less intuitive and the outputs of this approach were only used to compare with the hierarchical
classification and assist in identifying possible misclassifications.
3.4.6 Interpretation and refinement of classification
The outputs of the analyses outlined above were then interpreted and refined to produce a
draft classification scheme to present to the expert review panel. To support the interpretation
of the results dendrograms were produced at both the 300 and 150 group levels, allowing an
assessment of relationships between groups (see Appendix 4). Species fidelity statistics were
also produced following the approach described by Keith and Bedward (1999) to assist in
interpreting and defining groups and identifying possible misclassified sites.
Various environmental data were also analysed to support the interpretation of the classification
outputs. Site locations were loaded into a GIS to allow the spatial distribution of sites within each
group to be assessed and possible geographic outliers to be identified. A range of environmental
parameters was also captured in a GIS for each site, allowing this information to be summarised
for each proposed group. Parameters captured included topography (altitude, slope and
aspect), climate (mean annual temperature and rainfall), geology and lithology. In addition to
33