Discussion

4.4.1.Drainage analysis procedures

There is growing recognition of the value of consistent, continent-wide delineation of

hydrologic units (Bertolo, 2000; Franken et al., 2001; U.S. Geological Survey, 2001). Although developed for Australian drainage systems, the new drainage analysis procedures are generic and applicable anywhere that basic topographic mapping is available. They overcome many of the limitations of existing methods for continental scale application, effectively accounting for distributary and uncoordinated drainage systems and accurately delineating stream networks, even in regions of low relief.

These procedures recognise DEM development as a fundamental step in the drainage analysis. The accuracy of the drainage analysis products depends on the ability of the DEM to correctly represent surface drainage characteristics and is thus reliant on selection of appropriate DEM data sources and interpolation methods, and careful attention to error correction. Additional investment in DEM development, moreover, largely removes the need for DEM pre-processing routines such as stream burning and filling. Such pre-processing can be a significant and time- consuming undertaking. The Elevation Derivatives for National Applications (EDNA) project in the United States (Verdin and Greenlee, 2003), for example, oversees a major cooperative program to identify and correct errors and inadequacies in the national DEM (Gesch et al., 2002) to improve hydrological derivatives (Kost et al., n.d.). The production of a high quality

4.4 Discussion 78 DEM has the added benefit of supporting other applications requiring terrain derivatives (e.g.

landscape classification, see Chapter 7).

Setting appropriate thresholds for channel initiation is always challenging, but especially so at continental scales (Section 3.2.1.1). The solution developed here makes use of commonly available information in the form of cartographic streamlines. It does not attempt to define the true extent of the stream network, accepting that it cannot be reasonably determined from continental scale DEMs. Furthermore, the interface between hillslope and channel is highly dynamic (Kim and Lee, 2004). “Growing” the channel network from designated “start” cells is not a new idea (see for example review in Bertolo, 2000) but, surprisingly, the mapped channel heads do not appear to have been used to initiate the network.

The new drainage analysis procedures represent the major features of the drainage network without attempting to resolve the complex patterns of braided and anastomosed systems. For continental scale conservation planning it is not necessary (nor feasible) to model the complex fluvial and erosional processes (Makaske, 2001; Fagan and Nanson, 2004) responsible for the evolution of multi-channel planforms. Representing only major flow bifurcations avoids the problems of looping flow paths and other anomalies that may be encountered when multiple flow directions are included at every bifurcating node identified in the mapped streamlines (Stein et al., 2002). The number of flow bifurcations incorporated into the stream network is controlled by the value of the area threshold used to delineate sub-catchments. A reasonable threshold value was determined by trial and error from visual inspection of the resulting stream networks for selected test areas. Recognising flow bifurcations only where the branching stream arcs flowed to different terminal cells (inland sinks or coastal outlets) might be a more robust approach but would exclude distributaries within a drainage basin (e.g. the Cuttaburra/ Kulkyne Creek system that links the Paroo and Warrego Rivers, in the Murray-Darling Basin). The specification and operation of the criteria used to identify major bifurcations would benefit from further research.

Stream segments are bounded by tributary confluences. Ideally, stream segments are also broken at natural discontinuities such as major breaks in slope (Frissell et al., 1986) but developing a reliable method was not feasible for this study. A simple algorithm that breaks segments at grid cells overlayed by a cliff line and with a large drop in elevation from one cell to the next in the flow direction was tested, but requires further development. Many of the problems encountered could be attributed to the relatively coarse grid cell resolution so, for example, cliffs that were close to a confluence were attributed to the wrong stream segment and segments were erroneously divided where cliff lines ran close to, but parallel with, the stream (i.e. they delineated a gorge).

4.4 Discussion 79

4.4.2.Drainage analysis of the Australian DEM

The drainage analysis products that are the outcome of the application of the new procedures reflect the natural surface drainage characteristics of the Australian continent. Application of objective, systematically applied criteria produced a consistent representation of drainage networks without the confounding influences of variable cartographic interpretation. The geomorphometric properties of the derived stream network generally conform to the laws of drainage basin geometry usually ascribed to natural stream networks, supporting a contention that the intrinsic organization of river systems is preserved in the DEM-derived network (Vörösmarty et al., 2000).

Values for the Horton ratios typically fall within the range 3 to 5, 1.5 to 3 and 3 to 6 for the bifurcation (RB), stream length (RL) and drainage area (RA) ratios respectively (Kirchner, 1993).

The ratio values for the DEM-derived stream network (RB =3.5, RL =1.4, RA =3) are thus

towards the lower end or in the case of RL just outsideof the usual range. Lower values are

largely explained by the incorporation of uncoordinated and distributary drainage structures. Mean stream lengths will be lower than expected as a result of the breaks inserted into the DEM-derived network that correspond with gaps in the mapped streamline network and

difficulties assigning stream orders to distributary channel networks. The tests used to check for bifurcating channels re-joining, based on the elevation of the channel source, failed in certain circumstances, causing stream orders to be falsely incremented. Downstream of the confluence between an anabranch and a tributary originating from a higher source, the anabranch will be attributed as being derived from a new source (i.e. that of the tributary). As a consequence, stream order will be incremented if this anabranch rejoins the main stream further downstream, a situation demonstrated by the Flinders River in the Gulf of Carpentaria, the only river of order eleven. Inflated stream orders assigned to some distributary rivers will also cause the mean drainage area to be lower than is characteristic for a stream order. The single continent-wide value of the Horton ratios reported here may also be masking considerable variation between regions or drainage basins that would be usefully explored in the future.

4.4.3.Validation

The accuracy assessment undertaken in this study indicates that the new stream and catchment delineations will provide a reliable basis for continental scale conservation planning. The assessment corroborates the view of Hutchinson et al. (2000) that, at a nominal scale of about 1:250,000, the true surface drainage structure of the Australian continent is reliably determined from the national 9 second DEM. This scale is consistent with that recommended by the scientific guidelines developed for the National Reserve System Program of Australia

4.4 Discussion 80 (Commonwealth of Australia, 1999) and is appropriate for regional to continental scale

planning.

The buffer analysis and visual inspection of stream placements demonstrates the ability of the ANUDEM program to accurately represent drainage structures in the DEM, even in areas of low relief. Therefore, it avoids the need for stream burning or other DEM pre-processing algorithms. In most cases, divergence from the mapped streamlines can be attributed to the generalization of the input source data to the grid resolution by the ANUDEM program. Version 2 of the GEODATA TOPO –250K database (Geoscience Australia, 2003c) incorporates recent revisions to streamline mapping and potentially, therefore, locates streamlines more accurately. The version 2 database is available for free download. However, comprehensive coverage was only available from early in 2004. This was too late for this project. Disappointingly, the streamline vectors in version 2 do not retain the correct directions established for the version 1 streamline vectors. Furthermore, the connectors, added to bridge natural gaps in the drainage network, were poorly located, often taking little account of the underlying topography. In their current form then, the GEODATA version 2 streams are unsuitable for drainage analysis and DEM development.

The results of the buffer analysis compare favourably with those reported by Vogt, Colombo, Paracchini et al.(2003) for a pan-European stream and catchment dataset derived from a 250m resolution DEM. Despite the application of adaptive drainage enforcement algorithms (Soille et al., 2003) to improve the accuracy of the DEM flow paths in flat areas, the European stream network fell within a 250m buffer of the 1:50,000 scale reference streams for only 50% of its length, or within a 500m buffer for 87% of its length. In comparison, for all but the ACT 1:10,000 scale stream mapping, more than 96% of the grid cells of the stream network derived in this study were coincident with the rasterised reference stream network, indicating their locations differed by not more than about 375m. The highly variable mapping of drainage extent on the ACT 1:10,000 scale map series confounded comparison with the derived stream network. The accuracy assessment may, therefore, have been improved by considering only selected streamlines common to both the derived and reference datasets.

The buffer analysis of catchment boundaries showed that the drainage analysis of the Australian DEM also more accurately delineated catchment boundaries than the comparable European analysis. Less than 85% of the length of the derived catchment boundaries in the European dataset fell within a 1000m buffer around the reference catchment boundaries (Vogt, Colombo, Paracchini et al., 2003). In contrast, in this study more than 95% of the length of the derived catchment boundaries was within 500m of the reference set. This figure was almost 100% for the higher resolution reference data sets defining the Cotter Catchment boundary.

Summary statistics can, however, hide localized elevation errors that potentially have a very large influence on derived drainage networks and catchment areas (Jenson, 1991; Walker and

4.4 Discussion 81 Willgoose, 1999). While the likelihood of such effects is substantially reduced by the

comprehensive error correction procedures that were integral to DEM development and by the use of flow directions directly coded from the mapped streamlines, such errors need to be considered as a potential source of uncertainty in conservation planning applications based on the derived drainage network.