The oceanographic regionalisation was conducted by Dr Lee Belbin of the Australian Antarctic Division using a multivariate pattern classification system. Attributes were selected to be as independent as possible, and to maximise the information content of relevance to the regionalisation.
The analysis was designed to capture the seasonal patterns as well as the depth structure. Given these extra two dimensions, it was imperative to reduce the number of attributes down to a select few of high information value. The selection strategy was to represent the seasonal pattern with 4 layers chosen to represent the summer, autumn, winter and spring periods (the chosen dates are detailed in the covering notes on the maps). For the depth structure, a surface layer was required; to smooth some of the high variability, the layers from 0 - 50m were averaged. A second layer at 150m was chosen to depict a horizontal slice through the thermocline layer showing the intensity of penetration for the seasonal mixing signal. Finally a third layer averaged over the 800 - 1000m depth interval was chosen to represent such patterns as the extent of the deep salinity maximum - one of the deep water characteristics thought to influence distributions of deep-sea fauna.
As to the choice of attributes, temperature was an obvious one. Salinity was also chosen as the relationship between temperature and salinity varies both latitudinally and depthwise in response to changes in air-sea interactions processes as well as horizontal and vertical mixing. Hence the combination of these provided a good basis for the physical process effects. For the chemical and biological processes, nitrate was chosen to represent both nitrate and phosphate as well as the aspects of the functioning of the primary productivity cycles. Silicate was chosen over dissolved oxygen mainly for the deeper layer representation. In order to use dissolved oxygen, it has to be converted to percentage saturation to provide a meaningful attribute for the biological processes. One of the features/problems with silicate is that a component of its signal is associated with land-based processes. This became evident in the signal from islands such as New Calendonia where deep stations close to shore displayed high silicate signals.
Looking at the series of temperature maps, there is less structure and seasonality with depth. In the surface layer, the southward curvature of isotherms in the east and west is associated with the East Australian Current and the Leeuwin Current respectively. The warmest waters appear in the north east quadrant. This is more evident in the 150m plots which show a large warm pool east of Papua New Guinea. The western warm pool is at a more southern latitude and does not appear to be as intense. In the west the "upwelling" of cooler waters close to the coast of Java is prominent. There is less seasonal variability at the 150m layer and even less (of the same scale) in the deep layer. The deepest layer shows the coolest waters in the sub-Antarctic front region. The warmest deep water occurs in a number of tongues off eastern Australia, in the south western quadrant and off the west of Java.
The seasonal variability in the salinity signal is relatively weak except for the Indonesian Archipelago and the area east of Papua New Guinea. The disparity in the patterns in the east and west are however striking with remarkably large gradients in the west providing a stark contrast to the relatively homogeneous pattern in the east. The southern pattern is semi-continuous and does not display the same level of east-west contrast.
The nitrate pattern also shows a similar east-west disparity in pattern as salinity in the intermediate and deep layers; the surface layer is virtually stripped of nutrient expect for the strong gradient signal near the sub-Antarctic front. Large subsurface signals are also evident to the west of Java and Sumatra. The sparse number of stations in the deepest layer may be causing patterns to be biased by sampling.
Silicates in the surface display a relatively noisy signal with strong signals located in areas of islands and over subsurface topographic features such as the east Tasman Rise south of Tasmania. A very strong signal is evident in the 150m layer emanating from the north west of Australia and penetrating through to Java. This pattern expands in the deeper 800 - 1000m layer with very strong gradients located in a zonal band west of central Western Australian.
Looking through the other attribute maps, the 4 selected for the regionalisation attempt capture much of the range of the patterns of variability evident.
The regionalisations from these selected attributes are displayed in the appendix of maps and comprise a set of regionalisations for each of the 3 layers: 0 - 50m, 150m and 800 - 1000m. Regionalisations were conducted using groups of 10, 20 and 30. Out of these, the top layer and 150m appear to be well represented by the 20 group regionalisation, and the deep layer was well represented by the 10 group regionalisation.
It is remarkable how much of the patterns of attribute variability are captured in the regionalisations. The surface regionalisation shows the stronger western gradient grouping and the semi-continuous southern groups which have a slight southward dip proceeding eastward. The Indonesian Archipelago stands out as a unique environment in keeping with the patterns evident from the attributes. Similarly, the southward dip of isolines on the western and eastern coasts, associated with the boundary currents there, is also evident. Spencer Gulf in South Australia stands out as a distinct group reflecting the unique high temperature, and high salinity environment of that region. The 150m regionalisation provides the strongest reflection of the extraordinary east-west gradient disparity again in keeping with the salinity, nitrate and silicate signals. Here again the north western region stands out as a distinct pattern. The boundary current regions also appear to be better differentiated in the 150m pattern. In the deepest layer, there is distinct change in the east-west water masses across the north. Sharp changes are noted in a band running westward and slightly northward off central Western Australia with less variability south of this band.
Inspection of the property-property plots and the attribute range within each group shows that the pattern analysis has by and large differentiated the groups based on distinct changes in attribute values. The spatial patterns are also highly contiguous indicating that the analysis is reasonable.
Next Chapter: 16. Discussion