• No results found

The need for consistency dictated the use of spatial data layers with continental coverage. Unfortunately, the only data available for some attributes was sourced from small (coarse) scale mapping that may not support the 1:250,000 scale of the underlying spatial framework. This is particularly so for the geology mapping used. Produced at a scale of 1:2,500,000 the boundaries of mapping units on the Geology of Australia map are highly generalised. Map units represent the dominant lithological type yet minor geological formations may play an important role in shaping stream properties. The weathering products of sedimentary rocks, for instance, have a disproportionate influence on stream chemistry (Meybeck, 1987 cited in Knighton, 1998). While geological mapping in electronic form is available for much of the country at 1:250,000 scale or finer it would be a major undertaking to rectify the many inconsistencies that exist in the coding and lithological description of map units across state or map sheet tile boundaries. The NLWRA recently compiled the best available geological mapping into a consistently classified coverage, assigning map units to one of 23 categories of significance for soil

formation (National Land and Water Resources Audit, 2001e), but this task extended only to the Intensive Landuse Zone. Even so, there remain obvious coding discrepancies at State or map tile borders. A new national geological map is currently under development by Geoscience Australia at 1:1 million scale. As of September 2005, coverage of outcrop geology for the

7.4 Discussion 180 eastern states of Australia was available

(http://www.ga.gov.au/ausgeonews/ausgeonews200509/productnews.jsp).

Uncertainty also results from the overlay of data of different spatial scales (Bailey, 1988; Gaeuman et al., 2005). Combining the coarse scale geological mapping (1:2.5 million scale)

with much finer scale (1:250,000) climatic and topographic data will have produced spurious combinations of attributes and artificial groupings that have no ecological significance. To detect such artefacts, independent higher resolution data sources or field surveys would be required. Once available, the new seamless geological coverage at 1:1 million scale would undoubtedly improve future revisions of the river environment classifications, but scale mismatch, though less extreme, will continue to confound classifications. Whether the

information gains that arise from the inclusion of lithological attributes outweigh the increased uncertainty they produce is an important research question to follow up.

7.4.3.3. Number of groups

The number of groups in the classification, and hence the number of stream types recognised, will influence conservation planning outcomes including evaluation of the representativeness of existing protected areas (Pressey and Logan, 1994). More homogenous groups derive from a finer partitioning and are likely to provide better targets for representation of biodiversity in a protected area system (Pressey and Logan, 1994), a more accurate description of reference conditions for assessment of ecological condition and a reliable basis for extrapolation of survey results. Yet, too many groups will complicate management prescriptions and be more difficult to interpret and communicate.

Uncertainty in the attribute data limits much finer partitioning of stream segments in the ALL classification. At 355 groups, the differences in attribute values between the closest groups are approaching their estimated error (Table 7.14). This is clearly not the case, however, for the 10 group climate, flow and terrain classifications. Both the plots of average deviation and the modified classification strength indicate a finer partitioning would substantially reduce the heterogeneity of the resulting groups though neither readily identifies the best classification. A good classification reflects natural structure in the data (Bao, 2000) but natural partitions are not always readily apparent. While breaks coinciding with sudden changes in river character and behaviour at confluences (Benda et al., 2004) or abrupt changes in valley confinement (Brierley

and Fryirs, 2002) may be easily recognised, they are less perceptible when variation occurs continuously, reflecting change that is more gradual (Brierley and Fryirs, 2002).

There are no widely accepted criteria to objectively determine an optimal number of clusters (Belbin, 1993a). Furthermore, different indicators may not agree on this number (Mufti et al.,

7.4 Discussion 181 analysis of alternative partitions of the climate, flow and terrain classifications (Section 7.3.3), it

appears that a number of clusters between 20 and 30 might better balance the desire for within group homogeneity with the need for parsimony. Still, the lack of independence among these factors (each being a derivative of the DEM in some form) complicates any decision. Though more complex, analysing the adequacy of the level of partitioning of the classifications in combination would enable more robust conclusions.

7.5.Conclusions

For the first time, the rivers and streams of Australia have been consistently characterised continent-wide, producing a comprehensive picture of variation in landscape controls on aquatic ecosystem patterns and processes, both within and between drainage basins. The

classifications of stream segments summarise these patterns at multiple spatial scales, providing a potentially useful landscape framework for many conservation planning applications. First however, their ecological relevance must be validated (Chapter 8).

The classifications group hydrologic units according to their similarity in environmental data space. Consequently, group members may be widely dispersed geographically. The attributes selected to characterise landscape controls on stream systems reflect the important linkages between the stream and its catchment. Classes derive from the structure within the attribute data, rather than being imposed a priori.

Outcomes depend on the classification approach adopted. The environmental domain approach classified stream segments more parsimoniously (i.e. fewer groups) but groups tended to be less homogenous and therefore more difficult to interpret than the hierarchical classification. The number of types is an important property of a classification with implications for conservation planning applications. For a larger number of River Environment Types more rivers may be required to establish a representative system of protected rivers (Chapter 9). Acquiring sufficiently representative survey data to evaluate the ecological significance of the landscape classifications also becomes more difficult with increasing numbers of river types (Chapter 8).

8.1.1 Test data

8.1.1.

Test data

183

Chapter 8.

Are the landscape classifications