Data analysis

For the 1997 classification, data analyses using the TWINSPAN and DECORANA clustering and ordination techniques were employed to help define the types. The analytical processes adopted are described in Mills (1994).
 
The 1997 version was revised and refined to develop the present version. Extensive re-analyses of the data were carried out using  the analytical techniques available in PRIMER (Clarke & Warwick, 2001). The data were initially divided into the five broad habitat types shown in the primary habitat matrix, i.e. Littoral Rock, Littoral Sediment, Infralittoral Rock, Circalittoral Rock and Sublittoral Sediment. Due to the large size of the datasets within each broad habitat, some further a priori divisions of the data within broad habitats were necessary before analysis was possible. Additional analyses were carried out on data from "borderline" habitats to ensure these a priori splits did not force artificial divisions into the classification where this was not supported by differences in the survey data. Analysis within each broad habitat was led by a specialist for that habitat type. Figure 2 shows the data analysis process for the littoral sediment section. The following paragraphs describe the analyses carried out within each broad habitat:
 

Littoral rock

As the biotopes defined in version 97.06 (Connor et al., 1997 a, b) were generally considered satisfactory, analysis focused on clarifying the boundaries between closely related types and confirming the validity of certain less-well defined types. This included attention to the inter-relationship of fucoid-dominated types regarding the bedrock/boulder/mixed substrata and fully marine/variable salinity transitions and examination of the various red algal-dominated types. Additionally new data from intertidal caves enabled substantial development of the classification here. On the basis of these analyses, some restructuring at biotope complex level was necessary.
 

Littoral sediment

Due to the size of the Littoral Sediment dataset (>4000 records), some a priori division was necessary to provide datasets that could be managed within PRIMER (Clarke & Warwick, 2001). Data were divided based on the sediment type categories at habitat complex level in the 97.06 classification (Connor et al., 1997a, b): gravels and sands, muddy sands, sandy muds, muds and mixed sediments. Semi-quantitative epifaunal data were considered to be of less value than quantitative infaunal data for the purposes of the analysis and were thus excluded. Epifaunal data were however used to define types where a significant proportion of species would be sampled in epibiota sampling techniques, and/or where few infaunal samples were available, e.g. for mussel beds.
 
Cluster analysis was carried out based on species matrices listing individual counts per m2 in each sample, using the PRIMER software package (Clarke & Warwick, 2001). The data were divided into small clusters of biologically similar records, based on the resulting dendrograms. Comparative tables were produced to compare the species data and physical data between each of the small clusters. Where there were no notable differences between the physical and biological characteristics of the small clusters, they were amalgamated into larger groups which would form the preliminary basis for biotopes and sub-biotopes. Where similar biological and physical profiles appeared from clusters derived from different datasets, those data were joined and re-analysed. In particular, there was some overlap between the 'gravels and sands' and the 'muddy sands', and between the 'muddy sands' and 'mud' datasets. This re-analysis was carried out to ensure that the a priori divisions of the data did not artificially force divisions of otherwise coherent clusters. The resulting preliminary biotope and sub-biotope groups of records were then checked to ensure cohesion of both the environmental and species data. Individual records which differed significantly from the average profile for the group (in terms of biology or physical habitat characteristics) were removed, resulting in a group of records which formed the basis of the biotope descriptions (core biotope records). The physical and biological profiles from the core biotope records were then used to group biotopes of similar character into biotope complexes, and these in turn were assigned to habitat complexes and broad habitats. Note that, in addition to the habitat complexes defined on sediment character, two additional categories were created based on epifaunal characteristics (littoral sediments dominated by macrophytes, and littoral biogenic reefs).  
 

Infralittoral rock

As the biotopes defined in version 97.06 were generally considered satisfactory, analysis focused on clarifying the boundaries between closely related types and confirming the validity of certain less-well defined types. This included particular attention to the tide-swept kelp types and the inter-relationship of highly grazed and poorly grazed kelp habitats. On the basis of these analyses, some restructuring at biotope complex level was necessary.  Attention was also paid to the vertical rock section of the infralittoral rock classification, and examining how these additional biotopes could be fitted into the existing biotope complexes, reflecting the subtle differences in their biological character.
 

Circalittoral rock

Due to the complexities of this part of the classification, especially the more subtle differences between types on the open coast, a full re-analysis of the data were undertaken.  The large size of the circalittoral rock dataset meant that some a priori division was necessary to provide datasets that could be managed within PRIMER (Clarke & Warwick, 2001). Data were divided on the basis of three previously determined energy levels; high, moderate and low energy.  Cluster analysis was carried out using epifaunal species matrices exported from the AREV database, using the PRIMER software package (Clarke & Warwick, 2001). The data were divided into small clusters of biologically similar records, based on the resulting dendrograms. Comparative tables were produced to compare the species data and physical data between each of the small clusters. Where there were no notable differences between the physical and biological characteristics of the small clusters, they were amalgamated into larger groups which would form the preliminary basis for biotopes and sub-biotopes. Where similar biological and physical profiles appeared from clusters derived from different datasets, those data were joined and re-analysed.  This re-analysis was carried out to ensure that the a priori divisions of the data did not artificially force divisions of otherwise coherent clusters. The resulting preliminary biotope and sub-biotope groups of records were then checked to ensure cohesion of both the environmental and species data. Individual records which differed significantly from the average profile for the group (in terms of biology or physical habitat characteristics) were removed, resulting in a group of records which formed the basis of the biotope descriptions (core biotope records). The physical and biological profiles from the core biotope records were then used to group biotopes of similar character into biotope complexes, and these in turn were assigned to habitat complexes and broad habitats.  As in the infralittoral rock section, further analysis was also carried out on the vertical rock
section of the circalittoral rock classification.
 

Sublittoral sediment

A full re-analysis of the existing data on the MNCR database in addition to data supplied by the sublittoral specialist was carried out (approximately 10,000 records in total). This followed a similar approach to that described for littoral sediment and as outlined in Figure 2. Data were split according to sediment type, data type (infaunal or epibiota) and sampling technique (where appropriate). Poor quality data was also removed prior to analysis for later manual assessment. Cluster analysis was undertaken using either PRIMER (as described for the littoral sediments) or TWINSPAN (following the guidelines in Mills, 1994). Clusters of biologically similar records were produced and assessed using comparative tables. Clusters with poor species definition or highly variable physical characteristics were further sub-divided until more homogenous groups were derived. Where similar biological and physical profiles appeared from clusters derived from different main habitat datasets those data were combined and re-analysed using the same clustering methods as described above in order ensure that the a priori divisions of the data did not bias the results of the analysis.
 
Where similar biological and physical profiles were found in clusters from datasets of differing sampling method or those with different types of data (e.g. epibiota or infauna) the groups were re-analysed where possible at a lower level of resolution (either presence-absence or on the MNCR SACFOR scale) using PRIMER or TWINSPAN such that the differences in data type were reduced. As for the littoral sediments the resulting groups were then checked for cohesion with regard the physical and biological data, and individual records assigned to the groups were checked against the profiles of the groups as a whole and re-assigned if necessary.
 
The physical and biological profiles from the core records for each type were then used to group types of similar character into the broader biotope complexes and these in turn were assigned to one of the six main habitats for sublittoral sediment, derived from the EUNIS classification. The relationship between the sublittoral sediment biotopes is shown for separate depth bands in a series of habitat matrices, available to download as images from the classification website.