Printer Friendly

Construction and analysis of Hydrogeological Landscape units using Self-Organising Maps.

Introduction

The Hydrogeological Landscape (HGL) framework (Muller et al. 2015; Wooldridge et al. 2015) enables partitioning of the landscape into discrete regions or units. Each HGL unit spatially defines an area of similar landscape character, surface and groundwater flow pathways that is different from the surrounding landscape. The HGL framework methodology relies upon a team approach covering several different disciplines and skills including geology, geomorphology, pedology, hydrology, botany, land resource planning, government extension and local land management.

Development of HGL units is based on available existing data and integration of new field data. Standard data used for HGL unit development includes geology, regolith, soils, rainfall, land use, stream networks, waterbodies and vegetation mapping, digital elevation models (DEM), slope class analysis, water quality, groundwater and bore log data. Where available, gamma-ray spectroscopy (radiometrics), aerial photography, light detection and ranging (LiDAR), and electromagnetic (EM) induction surveys are also used. Models and indices used can include the weathering intensity index, which is derived from airborne gamma-ray spectrometry and digital terrain analysis (Wilford 2012), landform modelling (Summerell et al. 2005), multi-resolution index of valley bottom flatness (Gallant and Dowling 2003) and Fuzzy Landscape Analysis GIS topographic wetness modelling (Dowling et al. 2003; Summerell et al. 2004).

In this study, processing of spatial data and the manual interpretation of HGL units was achieved by using ArcGIS version 9.3 software (Environmental Systems Research Institute 2008). In general, a first-pass of data for defining HGL units would typically discern larger scale landform patterns (National Committee on Soil and Terrain 2009) by analysing the interplay of the typical relief from a DEM and modal terrain slope class. Mean annual rainfall and either geology or soil landscape mapping for the area (dependent on availability and scale of application) are used to determine similarities/ dissimilarities in physical materials, and geomorphic and hydrological characteristics and processes. All other data would then be used to refine HGL units until they arc appropriate to the size and scale of application. Typically, if an area of land is found to have two or more characteristic differences, it will be split into two different HGL units. Primary field observations and data are used to test and validate HGL unit boundaries and concepts and aid in the development of schematic cross-section diagrams for each HGL unit, showing the landscape features and processes that make it unique.

Self-Organising Maps (SOM; Kohonen 1982, 2001) is an unsupervised, statistical learning algorithm. SOM is a spatially constrained form of k-means clustering (Ripley 1996) that employs vector quantisation and measures of vector similarity as a means of 'mapping' multivariate samples to cells (nodes) arranged in 2D space (Kohonen 2001; Klose 2006; Bacao et al. 2008). The topology of neighbouring nodes in 2D space indicates their relative similarities in multivariate space (Bacao et al. 2008). SOM communicates the dominant patterns and structures within data via a multivariate vector (code-vector) that summarises the characteristics of samples associated with a particular node (Wehrens and Buydens 2007; Bierlein et al. 2008). These dominant patterns can be visualised via plots of component planes, indicating the relative contribution of input variables to a given node.

Self-Organising Maps is a popular algorithm for solving real-world problems across a diverse range of disciplines ranging from robotics to linguistics (Kohonen 2001; Cottrell and Verleysen 2006). In recent years, SOM has seen increasing use in spatially constrained, geo-environmental applications (Ferentinou et al. 2008). For example, Bedini (2009, 2012) and Cameiro et al. (2012) employed SOM to generate geological maps from airborne hyperspectral and geophysical data. These maps were used to infer the spatial distribution of significant lithological units and alteration zones at the Earth's surface. Ferentinou et al. (2008) and Mokarram et al. (2014) used SOM as a tool to develop an understanding of landform formative processes by mapping and classifying alluvial fans and drainage network channels from a DEM and associated derivatives, i.e. elevation, slope, aspect and curvature. Ferentinou and Sakellariou (2005, 2007) used SOM to classify landslide hazard from rock mechanics and soil property indices, and Ley et al. (2011) successfully classified catchments from data representing climatic, hydrologic and morphologic indices by using hierarchically merged SOM nodes. By interrogating SOM code-vector attributes and their topological relationships, several distinct catchment groups with similar hydrological response behaviours were identified. Lin and Chen (2006) used SOM to define regions with similar rainfall characteristics from rain-gauge data. The resulting SOM clusters were evaluated with respect to their spatial heterogeneity, i.e. clusters formed groups in geographic space. As stressed by Lin and Chen (2006) and Fraser and Dickson (2007), the success of SOM for spatial geo-environmental clustering applications is subject to the resulting clusters forming spatially contiguous regions with homogeneous characteristics.

Integration of SOM to the HGL framework of Wooldridge et al. (2015) and Muller et al. (2015) has the potential to increase confidence in the assessment of developed HGL units, especially where field-based validation is limited. In this study, we construct spatially contiguous regions representing HGL units by using SOM in a region of south-eastern New South Wales (NSW), Australia. Input spatial data are typical of those used for manually interpreting HGL boundaries, presenting extract summaries of these data to provide a framework with which to assign meaningful HGL attributes. The resulting SOM-HGL units are compared with 65 manually interpreted HGL units (Nicholson et al. 2014; A. L. Cowood, unpubl. data), which were developed using the same core datasets. This comparison assesses both the spatial and attribute properties of SOMHGL units with respect to land-management decision-support systems.

Methods

Data

The study region covers an area of ~60 000 [km.sup.2] and is in southeastern NSW, Australia (Fig. 1). All data were resampled to a coincident grid with 500 m resolution and Lambert Equal Area coordinate system, resulting in 234326 pixels (samples). In total, 81 variables, representing both numeric and dummy variables of categorical data, were input in the SOM analysis. To prevent variables from contributing unequal importance to SOM construction, given different resolutions of original spatial data, all variables were normalised to range 0-1 (Kalteh et al. 2008).

Numeric variables

Four numeric variables were provided to SOM for the construction of HGL units (Fig. 2): DEM, slope (SLOPE), topographic wetness index (TWI), and precipitation (RAIN). These spatial data layers were acquired from the NSW Office of Environment and Heritage corporate dataset, except for TWI, which was acquired from CSIRO. The DEM (Office of Environment and Heritage 2011) was sourced as a raster dataset with 30 m cell resolution covering NSW. SLOPE (Office of Environment and Heritage 2013) was derived from this DEM by using ArcGIS 9.3 and Spatial Analyst and output with a 30 m cell resolution. TWI (Gallant and Austin 2012) was calculated as [log.sub.c] (specific catchment area [m.sup.2]/slope%) and estimates the relative wetness within a catchment. The 3" (~90m) TWI raster was derived by combining partial contributing area, computed from a hydrologically enforced DEM (DEM-H), and percent slope, computed from a smoothed DEM (DEM-S). Both DEMS and DEM-H are based on the 1" resolution Shuttle Radar Topography Mission (SRTM) data acquired by NASA in February 2000. RAIN was derived from the BIOCLIM annual precipitation raster dataset for NSW (Office of Environment and Heritage 2009), which is based on the 250 m (9") cell resolution Geoscience Australia DEM raster dataset and generated using the ANUCLIM Beta algorithm (Xu and Hutchinson 2011). These data were aggregated and resampled from their original resolutions by using the mean of combined cells and then normalised to range 0-1.

Categorical variables

Four categorical variables were used for the construction of SOM-HGL units (Fig. 3): lithological units (GEO), soil classes (SOIL), vegetation type (VEG), and land-use classes (LAND). All spatial data layers were acquired from the NSW Office of Environment and Heritage corporate datasets. GEO ArcGIS shapefile polygons represent the 1:1 000 000 scale Surface Geology of Australia (Raymond and Retter 2010). SOIL ArcGIS shapefile polygons represent soil types from the 1:2 000 000 scale Digital Atlas of Australian Soils (National Resource Information Centre 1991) linked to level 2 soil classification information documented in Northcote et al. (1960). VEG data represent the Vegetation Formations of NSW and they were obtained as a raster dataset with a 200 m cell resolution (Keith and Simpson 2010). LAND ArcGIS shapefile polygons represent mapped classes based on the 1:50 000 scale Australian Land Use and Management Classification (Office of Environment and Heritage NSW 2011).

GEO units were simplified from 386 stratigraphic units to 20 lithology classes covering the study region. SOIL categories were generalised from 78 classes to 30 classes based on the initial two-character code for each soil class, which is associated with soil order level 2 (Isbell 1996). The VEG layer contains 13 classes and the LAND layer contains 14 classes. A summary of the classes used as input for individual categorical layers is provided in Table 1. If required, categorical variables were converted to raster data with a 500 m cell resolution, using the polygon that intersects the centre of each cell. Dummy variables were generated, which associate 0 for samples (cells) not intersecting a given class and 1 for samples that do. The creation of dummy variables generated 77 variables representing individual classes across these categorical datasets.

Data processing

Data processing and SOM analysis was carried out using the open source R statistical programming language (version 3.0) available from the Comprehensive R Archive Network (R Core Team 2015). Additional visualisations and comparisons with reference data were performed using ArcGIS version 10.1 software (Environmental Systems Research Institute 2012).

Construction of SOM-HGL units

Cluster analysis is initiated by presenting an m x n matrix representing m rows of samples and n columns of variables X = ([x.sub.1], [x.sub.2], [x.sub.n]) [member of] [R.sup.n], where all samples contain valid (non-missing) data to SOM. Via an iterative two-stage process, SOM nodes are trained from randomly seeded reference (seed-) vectors, M. First, [M.sub.i] seed-vectors are shown to the network and compared with all X that fall within a predefined radius of assessment. The closest seed-vector, or best matching unit (node) [M.sub.c], is commonly defined by using a Euclidean distance metric:

[parallel] X - [M.sub.i] [parallel] = [square root of [n.summation over [(i=1)] ([x.sub.i] - [m.sub.i]).sup.2]] (1)

and deemed most similar to X according to:

[parallel] X - [M.sub.c] [parallel] = min {[parallel]X - [M.sub.i][parallel]} (2)

Second, the weights of Mc and its neighbouring M, within the search neighbourhood N(t) are adjusted to correspond more closely to the properties of [x.sub.i]. The learning rate [alpha](t) controls the rate of change of [M.sub.c] weights during the adjustment process. These steps are repeated while reducing N(t) and [alpha](t), where t represents a given iteration. In this way, [M.sub.i] become trained nodes with code-vectors summarising the characteristics of associated input samples (Fraser and Dickson 2007; Peeters et al. 2007; Bierlein et al. 2008).

In this study, we set SOM training parameters N(t) to 2/3 of the total range observed within variables with a linear decrease in [alpha](t) from 0.05 to 0.01 over 2000 iterations. We trained 196 SOM nodes (X dimension = 14 and Y dimension = 14), using all available samples and randomly generated seed-nodes. A map representing spatial distribution SOM nodes across the study region was constructed by linking SOM nodes to their associated samples in a GIS.

We used a hierarchical agglomerative dendrogram clustering method, similar to that proposed by Vesanto and Alhoniemi (2000) and Siponen et al. (2001), to merge SOM nodes. The hierarchical agglomerative dendrogram clustering method merges SOM nodes by aggregating the two closest nodes in variable space, as defined by their code-vectors, and then calculates the similarity between this group of merged nodes and the remaining nodes. Finally, merged nodes are recursively aggregated until all nodes are grouped (Ripley 1996; Boudaillier and Hebrail 1998). Ward's minimum variance method (Ward 1963) was used to assess SOM node similarity because it attempts to find compact spherical groups of inputs.

An optimal number of merged SOM nodes (i.e. clusters) was identified using the Davies-Bouldin Index (DBI; Davies and Bouldin 1979):

DBI = 1/C [C.summation over (i=1)] [max.sub.j] ([S.sub.i] + [S.sub.j]/[M.sub.ij]) (3)

where C is the number of clusters, [S.sub.i] and [S.sub.j] are the mean squared distances from the centre of clusters i and j to all samples in clusters i and j, and [M.sub.ij] is the distance between the centroids of clusters i and j. Lower DBI values are obtained for clusters that are more internally compact and widely separated (Siponen et al. 2001), thus, indicating increasingly optimal representations of the natural groups within data.

For this study, an appropriate number of clusters (i.e. SOM-HGL units) was identified by plotting DBI as a function of the number of merged SOM nodes. The minimum number of merged SOM nodes resulting in a consistently low DBI was deemed to represent an optimal number of SOM-HGL units. Samples linked to a given cluster were plotted as a raster dataset representing the spatial distribution of SOM-HGL units. SOM-HGL units were further smoothed in the spatial domain using a 5 x 5 moving window that assigned the modal SOM-HGL unit of neighbouring cells to the centre cell. Spatial filtering was conducted to eliminate any high frequency noise that may be present in the results. The smoothed SOM-HGL raster was converted to spatial polygons for visualisation and so attributes, based on merged SOM node code-vectors, representing estimates of SOM-HGL units properties could be appended and queried.

Classification of SOM-HGL units

We summarised SOM-HGL polygons with attribute information indicating the dominant code-vector characteristics and linked this information to SOM-HGL spatial polygons. Attributes for individual SOM-HGL units include: SOM-HGL index, the number of samples (pixels) associated with that SOM-HGL, and the number of nodes merged to generate the SOM-HGL. Based on the information contained within SOM-HGL (merged) code-vectors, the following attributes were constructed for the eight original input variables: top-ranked category (based on proportion of merged code-vectors), top-ranked category proportion, second-top-ranked category, second-top-ranked category proportion, and entropy. For numeric input variables (i.e. DEM, SLOPE, TWI and RAIN), ranked categories and proportions represent 20th percentile classes based on SOM node code-vector distributions (Table 2). Figure 4 shows example frequency-histogram distributions for numeric input variables overlayed with the 20th percentile SOM node code-vector classes. For categorical input variables (i.e. GEO, SOIL, VEG and LAND), ranked categories and proportions represent input classes associated with SOM-HGL code-vectors.

Entropy is a measure of the relative distribution of proportions for candidate categories (Goodchild et al. 1994) and is defined in a standardised form as:

entropy = [parallel] 1/[log.sub.e] c [summation] [p.sub.c] [log.sub.e] [p.sub.c][parallel] (4)

where c is the number of candidate categories and [p.sub.c] is a vector of the relative proportions (summing to 1) of candidate categories. If most SOM-HGL code-vectors for a given input variable are associated with a single category, then entropy is ~0. Alternatively, if many categories have equivalent proportions then entropy approaches 1. Hence, entropy is a measure of heterogeneity and aids interpretation of attribute uncertainty within SOM-HGL units.

Results

The resulting SOM component planes for numeric and selected categorical attributes are plotted in Fig. 5. An inverse relationship between SLOPE and TWI is observed. Similarly, many nodes with high DEM values are associated with low RAIN values. The right-hand side of the SOM map represents moderate DEM and low SLOPE characteristics. Component plane plots for selected categorical variables are shown as a single component plane for individual classes. Code-vector values of 1 indicate that the class in question is associated with a particular node. If maximum code-vectors arc <1 for a given class, that class does not contribute significantly to the separation of SOM nodes.

Dominant classes for GEO arc (predominantly Siluro-Devonian) igneous felsic intrusive, igneous felsic volcanic, (Adaminaby Group) marine sedimentary siliciclastic, sedimentary siliciclastic (includes Siluro-Devonian rift basin, Permo-Triassic Sydney Basin and Cenozoic sediments), and to a lesser extent igneous mafic volcanic (Cenozoic basalts) and quartz-rich arenite to rudite (includes Siluro-Devonian rift basin and Permo-Triassic Sydney Basin sediments). Several dominant SOIL classes have been identified, i.e. Fa, LL, Mb, Me, Mw, Pb, Tb, and Ub (Table 1), but with a large degree of mixing between classes as indicated by many code-vector values >0 and <1. In VEG, cleared vegetation and LAND grazing classes are strongly associated with each other and with low SLOPE and low RAIN. Conservation Area and Tree and Shrub cover LAND classes are concentrated in nodes on the left-hand side and centre of the SOM map, respectively. The latter is associated with the dry sclerophyll forest (shrubby subformation) and grassy woodland (VEG) classes and high SLOPE and high RAIN.

Figure 6 shows DBI plotted as a function of the number (2-190) of merged SOM nodes. There is an overall decrease in DBI with an increase in the number of merged nodes. Several kinks or steps indicating an abrupt decrease in DBI are observed. Significant decreases in DBI do not occur for greater than 76 merged SOM nodes; hence, this represents the minimum number of merged nodes with consistently low DBI (<0.5) and is deemed to be the optimal number of SOM-HGL units for this study.

A comparison of the spatial distribution (consolidated using a 5 x 5 mean spatial filter) of the 76 SOM-HGL units with the 65 manually interpreted HGL units is shown in Fig. 7. Summaries of the percent cover of SOM-HGL units for individual HGL units are presented in Table 3. A single SOM-HGL unit was, on average, present in 17.7 HGL units (min. = 1, max. = 53), equivalent to 5.646% (min. = 0.012%, max. = 95.833%) of the total area covered by a HGL unit. A single HGL unit was, on average, represented by 20.7 nodes (min. = 2, max. =42). On average, the top four nodes (ranked by % area covered) cover 78.559% of a HGL unit (min. = 43.836%, max = 100%).

Discussion

This section initially takes a closer look at selected manually interpreted HGL units with respect to intersecting SOM-HGL units. Comparisons are made both in the spatial domain and by using the attributes associated with intersecting SOM-HGL units. We then provide an example of how SOM-HGL units can be used within the HGL framework and inform land-management decision-support systems.

Analysis of SOM-HGL units

At first glance, there does not appear to be significant spatial coincidence between the SOM-HGL units and HGL units (Fig. 7). However, closer inspection reveals that many HGL boundaries are intersected by SOM-HGL boundaries and individual HGL units comprise multiple SOM-HGL units. The following sections discuss the relevance of SOM-HGL spatial and attribute characteristics to two HGL units in detail: Monaro and Windellama. These two HGL units represent regions that differ substantially with respect to the complexity of represented landscapes, and thus provide an opportunity to illustrate the potential for SOM-HGL units to capture similarities and dissimilarities in physical characteristics of contrasting landscapes. The Monaro HGL shows good spatial coincidence with two SOM-HGL units (10 in total) and it is considered to represent a relatively homogeneous landscape. By contrast, the Windellama HGL intersects a large number of SOM-HGL units (20 in total) and represents a heterogeneous and complex landscape.

Monaro HGL

The manually interpreted Monaro HGL unit is intersected by 10 SOM-HGL units, although units 38 (82.62%) and 8 (15.35%) account for 97.97% of this area, whereas remaining units, such as 72 and 32, contribute <1% each. The geometry of SOM-HGL 38 closely mimics the majority of Monaro boundaries (Fig. 8). However, this SOM-HGL extends to areas of Barneys Range, Alluvial, Murrumbucca and Brown Mountain HGLs to the north-west and south-east of Monaro. SOM-HGL 8 fills in small areas of the Monaro HGL in the north and west. SOM-HGL 72 is found adjacent to 38 and extends northwards from Monaro, and SOM-HGL 32 is a small area located within 38. The relationship of SOM-HGL 38 with Brown Mountain and Murrumbucca HGLs is significant because these HGL units were initially all part of the manually interpreted Monaro HGL. Brown Mountain HGL is indicative of higher relief and steeper sloped landform patterns on Ccnozoic basalt, whereas Murrumbucca HGL was separated out because it represents areas of shallower stacked basalt lava and a greater presence and influence from underlying lithologies. Similarly, the small Alluvial HGL units are geologically recent alluvial deposits overlying the older basalt lavas of the Monaro HGL unit.

The Monaro HGL exhibits a multi-cyclic erosional landscape, which typically forms an undulating to hilly dissected tableland with some rounded hills, flat-topped ridges and small valley plains (Tulau 1994; Nicholson et al. 2014). Table 4 summarises SOM-HGL attributes and shows that 38 and 8 are both associated with high to very high DEM, very low to low SLOPE, and high to very high TWI. The proportions of primary and secondary DEM, SLOPE and TWI categories are approximately equivalent for both SOM-HGLs as indicated by their similar entropy values. However, the higher entropy for SLOPE and TWI compared with DEM suggests a higher degree of variability in these inputs, potentially indicating undulating landforms. This HGL receives 500-750 mm mean annual rainfall (Office of Environment and Heritage 2009), which is apparent in the very low RAIN associated with all intersecting SOM-HGLs.

The Monaro HGL comprises jointed and layered Cenozoic basalt lava flows. The lava flows typically overlie Ordovician metasedimentary Adaminaby Group rocks or Siluro-Devonian granitic rocks, which can be exposed on lower slopes and in drainage lines (Raymond and Retter 2010; Nicholson et al. 2014). The SOM-HGL GEO attribute information supports this interpretation, indicating igneous mafic volcanic rocks dominate, with SOM-HGL 38 containing small areas of marine sedimentary siliciclastic rocks and SOM-HGL 8 containing minor igneous felsic intrusives.

Upper slopes and ridge tops form red and brown friable earths (Ferrosols) with friable neutral red soils in association with cracking clays on mid and lower slopes; also, there are some dark friable earths on lower slopes. In association with valley plains are various cracking clays. Hard neutral red soils occur on the crests of some hills whereas stony dark porous loamy soils occur on some lower hill slopes (Northcote et al. 1960; Tulau 1994). SOM-HGL units 38 and 8 are dominated by Md and to a lesser extent LL SOIL attributes, indicating Ferrosols and loamy soils, respectively.

The Monaro is extensively cleared with remaining vegetation assemblages comprising Grasslands with some areas of Grassy Woodlands and montane Freshwater Wetlands (Gellie 2005; Keith and Simpson 2010). SOM-HGL unit 38 is dominated by cleared vegetation, whereas SOM-HGL 8 represents a mixture of grasslands and cleared vegetation. Land use in the Monaro HGL is predominantly grazing. All intersecting SOM-HGL units in Table 4 are dominated by the grazing land-use category and contain minor tree and shrub cover.

Major differences between SOM-HGL units 38 (and 8) and 72 are observed in the GEO and SOIL attributes. SOM-HGL 72 is associated with sedimentary siliciclastic and minor regolith (alluvium) and hard neutral yellow and yellow mottled soils (Ub) and minor acid leached yellow earths (Mr). These contrasting properties, and the significant area covered by SOM-HGL 72 outside of the Monaro HGL, suggest that it represents a different HGL unit or would be better associated with a geologically transitional unit such as Murrumbucca HGL.

Windellama HGL

The manually interpreted Windellama HGL unit contains 20 SOM-HGL units, with units 16 (35.31%), 18 (19.59%), 37 (11.32%) and 24 (9.17%) accounting for 75.39% of this area (Fig. 9). These SOM-HGL units are also adjacent to each other within distinct regions to the north-west and south-east of Windellama, which suggests that neighbouring areas share the characteristics of this HGL. The Windellama HGL displays more complex landscape characteristics than the Monaro HGL unit. The reduced relationship fit with the SOM-HGLs is indicative of the internal complexity of this manually interpreted HGL unit.

The Windellama HGL typically forms hilly to steep hilly ranges, with rock outcrops, and ridge and valley terrain of mild relief (Jenkins 1996; Jenkins et al. 2010; Nicholson et al. 2014). The southern extent of this HGL contains undissected and dissected river terraces. The SOM-HGL units 16, 18, 37 and 24 DEM, SLOPE and TWI attributes in Table 4 are highly variable, as indicated by entropy values approaching 1. With respect to elevation, SOM-HGL 16 displays low to moderate values, SOM-HGL 37 and 24 high to moderate values, and SOM-HGL 18 high to very high values. These SOM-HGLs all exhibit very low to low SLOPE and high to very high TWI, except for 37, which is associated with very low TWI.

Windellama receives 600-750 mm mean annual rainfall (Office of Environment and Heritage 2009). The SOM-HGLs associated with this HGL unit have RAIN attributes that range from very low to moderate. However, these SOM-HGLs display widely varying RAIN characteristics as indicated by relatively high entropy values.

The Windellama HGL comprises Ordovician interbedded deep marine sandstones, siltstones, black shales and cherts, and incipiently metamorphosed equivalents of the Adaminaby and Bendoc Groups (Jenkins et al. 2010; Raymond and Retter 2010; Nicholson et al. 2014). SOM-HGL units that intersect this HGL are dominated by marine sedimentary siliciclastic rocks and minor non-marine sedimentary siliciclastic rocks (16 and 24) and igneous mafic volcanic rocks (18 and 37).

The chief soils for this HGL are shallow stony hard acidic yellow mottled soils (Tb) on the ranges and hard acidic yellow and yellow mottled soils (Ub) on low undulating and often stony ridges (Northcote et al. 1960; Jenkins 1996). SOM-HGL units 16, 18, 37 and 24 are dominated by Tb and to a lesser extent Ub SOIL attributes. Windellama has only minor clearing, and vegetation assemblages are dominated by Dry Sclerophyll Forest with minor assemblages of Forested Wetlands, Freshwater Wetlands and Grassy Woodlands (Keith and Simpson 2010; Tozer et al. 2010).

The SOM-HGL units intersecting Windellama distinguish hard and neutal to acidic (Tb) yellow and yellow mottled soils associated with dominantly dry sclerophyll forests on undulating ridges (16) and cleared land used for grazing on loamy and stony soils (LL) (18). Small areas of undulating ridges with mixed forest and cleared land are used for grazing (37). SOM-HGL 24 identifies areas on cleared land used for, or close to, rural residential zones. Although the SOM-HGL character is not as concise for Windellama, the interplay of the dominant geology, soil types and low rainfall across the present SOM-HGL units has a large influence on the land capabilities and limitations (Jenkins 1996). The manually interpreted HGL unit was delineated to represent the local high-risk land management areas instead of a more uniform landform pattern and character, which is shown in the SOM-HGLs.

SOM-HGL as a hydrological characterisation tool

Some of the data utilised in this study to construct SOM-HGL units representing regions with similar landscape characteristics are also commonly used to derive hydrological characteristics. For example, Coram (1998) and Coram et al. (2001) determined the hydrogeology of groundwater flow systems (GFS) through analysis of data analogous to those used in this study. The GFS has been widely used to inform salinity management in Australia (National Land and Water Resources Audit 2001; Walker et al. 2003; Dowling et al. 2004). This is also consistent with the methodology of Winter (2001) and Wolock et al. (2004), who have employed cluster analysis to delineate hydrologic landscapes for wetland management applications by using metrics of landform, geology, soil and climate. Currently, the HGL framework methodology (Muller et al. 2015; Wooldridge et al. 2015) uses the landscape characteristics present within HGL units to infer established hydrological properties and processes identified in the literature, including those mentioned above. To increase the functionality of determining hydrological characterisation in SOM-HGL units, a separate analysis, using a restricted set of data representing hydrological information, could be undertaken. In addition, spatial data appropriate to defining hydrological characteristics can be incorporated into the SOM-HGL analysis, such as the GFS and regolith weathering intensity index (Wilford 2012). Hence, the scope of potential applications of our SOM-based methodology is limited only by the selection of appropriate input data.

SOM-HGL as a management tool

The comparison of manually interpreted HGL units with our SOM-HGL units above highlights the validity of the SOM-based approach. We objectively defined spatially contiguous and homogeneous regions with distinct, comparable and meaningful attributes. However, the real value of SOM-HGL units will be realised when they contribute to land management decision-support systems, a primary objective of the HGL framework.

The HGL framework has tailored natural-resource-management components, which arc applied after HGL unit development. Management components facilitate land managers to undertake appropriate actions specific to individual HGL units and management areas (Wooldridge et al. 2015). Schematic cross-sections and landform modelling allow for the development of management areas (land within an HGL unit that can be managed in a uniform manner), which are determined by identifying landscape facets based on the Australian Soil and Land Survey Field Handbook (National Committee on Soil and Terrain 2009). Current management applications include dryland salinity, urban salinity, catchment action planning, revegetation, wetlands and climate change (Jenkins et al. 2010; Nicholson et al. 2011a, 2011b; Muller et al. 2012; Cowood et al. 2014; Nicholson et al. 2014).

As previously stated, SOM-HGLs intersecting the Windellama HGL (Fig. 9) grouped together in areas outside of this HGL, suggesting that these neighbouring regions share characteristics similar to the Windellama HGL. By executing attribute queries, we can identify which SOM-HGLs are linked in both geographical and attribute space. For example, the Boolean attribute query:

* GEO = 'igneous felsic intrusives'

* AND VEG = 'cleared'

* AND SOIL = 'Tb' OR 'Pb'

* AND RAIN = 'very low' OR 'low'

* AND SLOPE = 'very low' OR 'low'

selects three SOM-HGL units: 39, 56 and 57. Plots of these SOM-HGL units with respect to the dominant intersecting HGL units in geographical space are shown in Fig. 10. It is apparent that a very close correspondence exists between selected SOM-HGLs and several HGL units. Furthermore, these SOM-HGLs group together on the SOM 2D map (inset box Fig. 10), confirming their similarities in attribute space. The obvious similarities between these SOM-HGLs suggest that several HGL units share similar physical attributes and, thus, are likely require similar land management strategies.

Conclusions

Automated construction of HGL units using SOM from widely available spatial datasets shows great potential to influence the development process of manually interpreted HGL units. SOM-based statistical analysis provides a tool for the objective assessment of similarities/dissimilarities within and between multiple datasets indicative of landscape character. SOM-based methods will be especially useful over large and/or inaccessible regions, where conducting field-based validation is either logistically or economically impractical. The major drawback is that all data layers provided to SOM must cover the same geographic extent to avoid data gaps in the output.

Comparisons of SOM-derived HGL units with manually interpreted HGL units show good correlation between these data. Manual HGL unit interpretation accounts for a certain level of internal variation within a given HGL unit. By contrast, SOM-HGL units define added complexities, not apparent within individual HGL units, delineated by spatially contiguous subregions. These subregions represent subtle but important variations in geology, soil, vegetation, landscape morphology and climate.

We demonstrate how SOM-HGL units can also be used to identify, via attribute queries, boarder regions with similar physical characteristics that coincide with multiple HGL units. In contrast to the manually interpreted HGL units, which tend to be located within a single climatic zone and/or physiographic/bioregion, selected SOM-HGL units are not as confined to a specific geographical region. Thus, SOM-HGL units can be used to assign appropriate land management systems to broad regions that share physical characteristics. With appropriate input data, SOM can be used to complement expert knowledge and will contribute significantly to a range of land-management decision-support systems based on the HGL framework.

Acknowledgements

The authors acknowledge the R project for statistical computing (http:// www.r-project.org). This research was conducted, in part, using the High Performance Computing (HPC) facilities provided by the Tasmanian Partnership for Advanced Computing (TPAC). This research has used data that are [c] Commonwealth of Australia 2003 and have been used with the permission of Geoscience Australia and CSIRO. Limited End-user licence provided by Geoscience Australia and CSIRO. The application of the Hydrogeological Landscape framework detailed in this manuscript was done in conjunction with the Landscape Management Technical Group, a NSW Government cross-agency group consisting of Allan Nicholson, Andrew Wooldridge, Marion Winkler and Jan Wightley from the NSW Department of Primary Industries, Rob Muller, Brian Jenkins and Wayne Cook from the NSW Office of Environment and Heritage, and Leah Moore and Alison Cowood from the University of Canberra. NSW Environment Trust and Australian Government Regional Natural Resource Management Planning for Climate Change Fund provided funding through collaboration with Donna Hazell and Kristy Moyle of South East Local Land Services.

References

Bacao F, Lobo V, Painho M (2008) Introduction: What is a self-organzing map. In 'Applications of different self-organizing map variants to geographical information science problems'. (Eds A Skupin, P Agarwal) pp. 21-44. (John Wiley & Sons Ltd.: Chichester, UK)

Bedini E (2009) Mapping lithology of the Sarfartoq carbonatite complex, southern West Greenland, using HyMap imaging spectrometer data. Remote Sensing of Environment 113, 1208-1219. doi:10.1016/j.rse. 2009.02.007

Bedini E (2012) Mapping alteration minerals at Malmbjerg molybdenum deposit, central East Greenland, by Kohonen Self-Organizing Maps and matched filter analysis of HyMap data. International Journal of Remote Sensing 33, 939-961. doi: 10.1080/01431161.2010.542202

Bierlein FP, Fraser SJ, Brown WM, Lees T (2008) Advanced methodologies for the analysis of databases of mineral deposits and major faults. Australian Journal of Earth Sciences 55, 79-99. doi: 10.1080/08120090 701581406

Boudaillier E, Hebrail G (1998) Interactive interpretation of hierarchical clustering. Intelligent Data Analysis 2, 229-244. doi: 10.1016/S1088-467X(98)00026-2

Carneiro CC, Fraser SJ, Croacutesta AP, Silva AM, Barros CEM (2012) Semiautomated geologic mapping using self-organizing maps and airborne geophysics in the Brazilian Amazon. Geophysics 77, K17 K24. doi:10.1190/geo2011-0302.1

Coram J (1998) 'National classification of catchments for land and river salinity control: A catalogue of groundwater systems responsible for dryland salinity in Australia.' Publication No. 98/78. (Rural Industries Research and Development Corporation: Canberra, ACT)

Coram JE, Dyson PR, Evans WR (2001) An evaluation framework for dryland salinity. A report prepared for the National Land and Water Resources Audit Dryland Salinity Project, Bureau of Rural Sciences, Canberra, ACT.

Cottrell M, Verleysen M (2006) Special Issue: Advances in Self-Organizing Maps-WSOM'05. Neural Networks 19, 721-722. doi:10.1016/j.neunet. 2006.05.011

Cowood AL, Moore CL, Nicholson A, Hazell D (2014) Wetland conservation in the face of climate change: Summary report. NSW Office of Environment and Heritage, Sydney.

Davies DL, Bouldin DW (1979) A cluster separation measure. IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-1, 224-227. doi: 10.1109/TPAMI.1979.4766909

Dowling T, Summerell GK, Walker J (2003) Soil wetness as an indicator of stream salinity: a landscape position index approach. Applying Computer Research to Environmental Problems 18, 587-593. doi: 10.1016/S1364-8152(03)00034-3

Dowling T, Dawes W, Evans R, Dyson P, Walker G (2004) Prioritising upland catchments in the Murray-Darling Basin with respect to salinity benefits from afforestation. CSIRO Land and Water Technical Report No. 15/04. CSIRO, Canberra.

Environmental Systems Research Institute (2008) 'ArcGIS Desktop: Release 9.3.' (Environmental Systems Research Institute: Redlands, CA, USA)

Environmental Systems Research Institute (2012) 'ArcGIS Desktop: Release 10.1.' (Environmental Systems Research Institute: Redlands, CA, USA)

Ferentinou MD, Sakellariou MG (2005) Assessing landslide hazard on medium and large scales, using SOM. In 'Landslide risk management'. (Eds O Hungr, R Fell, R Couture, E Eberhardt) pp. 639-648. (Taylor & Francis Group: London)

Ferentinou MD, Sakellariou MG (2007) Computational intelligence tools for the prediction of slope performance. Computers and Geotechnics Special Issue on Biologically Inspired and Other Novel Computing Techniques in Geomechanics 34, 362-384. doi:10.1016/j.compgeo. 2007.06.004

Ferentinou M, Karymbalis E, Charou E, Sakellariou M (2008) Using self organising maps in applied geomorphology. In 'Self-organising maps: Applications in geographic information science'. (Eds A Skupin, P Agarwal) pp. 273-298. (Wiley: Chichester, UK)

Fraser SJ, Dickson BL (2007) A new method for data integration and integrated data interpretation: Self-organising maps. In '5th Decennial International Conference on Mineral Exploration'. Expanded Abstracts, pp. 907-910. (Decennial Mineral Exploration Conferences) Available at: www.dmec.ca/ex07-dvd/E07/pdfs/67.pdf

Gallant J, Austin J (2012) Topographic wetness index (3' resolution) derived from 1' SRTM DEM-H. v1. CSIRO Data Collection. Available at: https://data.csiro.au/dap/landingpage?pid=csiro%3A5588

Gallant JC, Dowling TI (2003) A multiresolution index of valley bottom Harness for mapping depositional areas. Water Resources Research 39, 1347. doi : 10.1029/2002WR001426

Gellie NJH (2005) Native vegetation of the Southern Forests: South-east Highlands, Australian Alps, South-west Slopes, and SE Corner bioregions. Cunninghamia 9, 219-254.

Goodchild MF, Chih-Chang L, Leung Y (1994) Visualizing fuzzy maps. In 'Visualization in GIS'. (Eds HM Hearnshaw, DJ Unwin) pp. 158-167. (Wiley: New York)

Isbell RF (1996) 'The Australian Soil Classification.' 1st edn (CSIRO Publishing: Melbourne)

Jenkins BR (1996) Soil landscapes of the Braidwood 1:100,000 sheet map and report. NSW Department of Land and Water Conservation, Sydney. Available at: www.environment.nsw.gov.au/soils/braidwood.htm

Jenkins BR, Nicholson A, Moore CL, Harvey K, Cook W, Wooldridge A, Shoemark V, Nowakowski A, Muller R, Winkler M (2010) Hydrogeological landscapes for the Southern Rivers Catchment Management Authority, Braidwood 1:100,000 map sheet. NSW Department of Environment, Climate Change and Water, Sydney.

Kalteh AM, Hjorth P, Berndtsson R (2008) Review of the self-organizing map (SOM) approach in water resources: Analysis, modelling and application. Environmental Modelling & Software 23, 835-845. doi: 10.1016/j.envsoft.2007.10.001

Keith DA, Simpson CC (2010) Vegetation formations of NSW (version 3.0): A seamless map for modelling fire spread and behaviour. Report to the Rural Fire Service. Version 2 data: NSWmap_v2_p1750_VIS_3846 & NSWmap_v2_ext_VIS_3847. NSW Department of Environment & Climate Change, Sydney.

Klose C (2006) Self-organizing maps for geoscientific data analysis: geological interpretation of multidimensional geophysical data. Computational Geosciences 10, 265-277. doi:10.1007/s10596-006-9022-x

Kohonen T (1982) Self-organized formation of topologically correct feature maps. Biological Cybernetics 43, 59-69. doi:10.1007/BF0033 7288

Kohonen T (2001) 'Self-organizing maps.' (Springer: Berlin)

Ley R, Casper MC, Hellebrand H, Merz R (2011) Catchment classification by runoff behaviour with self-organizing maps (SOM). Hydrology and Earth System Sciences 15, 2947-2962. doi: 10.5194/hess-15-2947-2011

Lin G-F, Chen L-H (2006) Identification of homogeneous regions for regional frequency analysis using the self-organizing map. Journal of Hydrology 324, 1-9. doi: 10.1016/j.jhydrol.2005.09.009

Mokarram M, Seif A, Sathyamoorthy D (2014) Use of morphometric analysis and self-organizing maps for alluvial fan classification: case study on Ostorankooh altitudes, Iran. In 'IOP Conference Series: Earth and Environmental Science'. Kuala Lumpur. 012003. (IOP Publishing) Available at: http://iopscience.iop.org/1755-1315/20/1/012003

Muller R, Jenkins BR, Cook W, Littleboy M, Nicholson A, Moore CL, Wooldridge A, Winkler M (2012) Salinity Hazard for Catchment Action Plan (CAP) Updates, Murrumbidgee Catchment Management Authority Area. NSW Department of Primary Industries/NSW Office of Environment and Heritage, Department of Premier and Cabinet, Wagga Wagga, NSW.

Muller R, Nicholson A, Wooldridge A, Jenkins BR, Winkler M, Cook W, Grant S, Moore L (2015) Hydrogeological Landscapes for the Murray Catchment Management Authority: Eastern Murray Catchment. NSW Office of Environment and Heritage, Sydney. Available at: www. environment.nsw.gov.au/publications/salinity/150240-HGL-Eastern-Murray.htm

National Committee on Soil and Terrain (2009) 'Australian soil and land survey field handbook.' (CSIRO Publishing: Melbourne)

National Land and Water Resources Audit (2001) 'Australian Dryland Salinity Assessment 2000. Extent, impacts, processes, monitoring and management options.' (Land & Water Australia: Canberra, ACT)

National Resource Information Centre (1991) The Digital Atlas of Australian Soils--scanned tracings. ASRIS, CSIRO. Available at: www.asris.csiro. au/themes/Atlas.html

Nicholson A, Cook W, Nowakowski A, Littleboy M, Ji F, Wightley J (2011a) Dubbo Urban Landscape Interpretation Project Final Report. Dubbo City Council Central West Catchment Management Authority. NSW Department of Environment, Climate Change and Water, Sydney.

Nicholson A, Pavan N, Winkler M, Taylor L, Cull V, Copley S, Muller R, Jenkins BR, Cook W, Shoemark V, Nowakowski A, Marchand A, Moore CL (20116) Hydrogeological Landscapes for the Sydney Metropolitan Catchment Management Authority, Western Study Area. NSW Department of Environment, Climate Change and Water, Sydney.

Nicholson A, Cowood A, Muller R, Wooldridge A, Cook W (2014) Impact of regional corridor plantings on water quality--South East Local Land Services. NSW Department of Primary Industries, Sydney.

Northcote KH, Beckmann GG, Bettenay E, Churchward HM, van Dijk DC, Dimmock GM, Hubble GD, Isbell RF, McArthur WM, Murtha GG, Nicolls KD, Paton TR, Thompson CH, Webb AA, Wright MJ (1960) 'Atlas of Australian Soils, Sheets 1 to 10 with explanatory data.' (CSIRO Australia and Melbourne University Press: Melbourne)

Office of Environment and Heritage (2009) BIOCLIM annual precipitation (digital dataset). Department of Premier and Cabinet, Canberra, ACT.

Office of Environment and Heritage (2011) DEM 30 m (digital dataset). Department of Premier and Cabinet, Canberra, ACT.

Office of Environment and Heritage (2013) Slope 30 m (degrees) (digital dataset). Department of Premier and Cabinet, Canberra, ACT.

Office of Environment and Heritage NSW (2011) Landuse. Department of Environment, Climate Change and Water, Sydney. Available at: www. agriculture.gov.au/abares/aclump/land-use/data-download

Peeters L, Lobo V, Dassargues A (2007) Exploratory data analysis and clustering of multivariate spatial hydrogeological data by means of GEO3DSOM, a variant of Kohonen's Self-Organizing Map. Hydrology and Earth System Sciences 11, 1309 1321. doi: 10.5194/ hess-11-1309-2007

R Core Team (2015) 'R: A Language and Environment for Statistical Computing.' (R Foundation for Statistical Computing: Vienna, Austria) Available at: www.R-project.org/

Raymond OL, Retter AJ (2010) Surface geology of Australia 1:1,000,000 scale, 2010 edn (digital dataset). Geoscience Australia, Commonwealth of Australia, Canberra, ACT. Available at: www.ga.gov.au

Ripley BD (1996) 'Pattern recognition and neural networks.' (Cambridge University Press: Cambridge, UK)

Siponen M, Vesanto J, Simula O, Vasara P (2001) An approach to automated interpretation of SOM. In 'Advances in self-organising maps'. (Eds N Allinson, H Yin, L Allinson, J Slack) pp. 89-94. (Springer: London)

Summerell GK, Dowling TI, Wild JA, Beale G (2004) FLAG UPNESS and its application for mapping seasonally wet to waterlogged soils. Soil Research 42, 155-162. doi:10.1071/SR03028

Summerell GK, Vaze J, Tuteja NK, Grayson RB, Beale G, Dowling TI (2005) Delineating the major landforms of catchments using an objective hydrological terrain analysis method. Water Resources Research 41, W12416. doi: 10.1029/2005WR004013

Tozer MG, Turner K, Keith DA, Tindall D, Pennay C, Simpson CC, MacKenzie B, Beukers P, Cox S (2010) Native vegetation of southeast NSW: a revised classification and map for the coast and eastern tablelands. Cunninghamia 11, 359-406.

Tulau MJ (1994) Soil landscapes of the Cooma 1:100,000 Sheet map and report. Department of Conservation and Land Management, Sydney. Available at: www.environment.nsw.gov.au/soils/cooma.htm

Vesanto J, Alhoniemi E (2000) Clustering of the self-organizing map. IEEE Transactions on Neural Networks 11, 586-600. doi: 10.1109/72.846731

Walker G, Gilfedder M, Evans R, Dyson P, Stuaffacher M (2003) Groundwater flow systems framework--essential tools for planning salinity management. Summary Report, Murray Darling Basin Commission. Land and Water Australia, Canberra, ACT. Available at: http://lwa.gov.au/products/pk030627

Ward JH (1963) Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association 58, 236-244. doi: 10.1080/01621459.1963.10500845

Wehrens R, Buydens LMC (2007) Self- and super-organising maps in R: the kohonen package. Journal of Statistical Software 21, 1-19.

Wilford J (2012) A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis. Geoderma 183-184, 124-142. doi:10.1016/j.geoderma.2010. 12.022

Winter TC (2001) The concept of hydrologic landscapes. Journal of the American Water Resources Association 37, 335-349. doi: 10.1111/ j.1752-1688.2001.tb00973.x

Wolock DM, Winter TC, McMahon G (2004) Delineation and evaluation of hydrologic-landscape regions in the United States using geographic information system tools and multivariate statistical analyses. Environmental Management 34(Suppl 1), S71-S88. doi: 10.1007/ S00267-003-5077-9

Wooldridge A, Nicholson A, Muller R, Jenkins BR, Wilford J, Winkler M (2015) Guidelines for managing salinity in rural areas. NSW Office of Environment and Heritage, Sydney. Available at: www.environment. nsw.gov.au/publications/salinity/150241-HGL-salinity-rural.htm

Xu T, Hutchinson M (2011) 'ANUCLIM version 6.1 user guide.' (Fenner School of Environment and Society, Australian National University: Canberra, ACT)

http://dx.doi.org/10.1071/SR15016

M. J. Cracknell, (A,B,E) and A. L. Cowood (C,D)

(A) School of Physical Sciences (Earth Sciences), University of Tasmania, Private Bag 79, Hobart, Tas. 7001 Australia.

(B) Centre of Excellence in Ore Deposits (CODES), University of Tasmania, Private Bag 79, Hobart Tas. 7001, Australia; Current address: ARC Research Hub for Transforming the Mining Value Chain, University of Tasmania, Private Bag 79, Hobart, Tas. 7001, Australia.

(C) Dryland Salinity Hazard Mitigation Program, University of Canberra, ACT 2601, Australia, institute for Applied Ecology, University of Canberra, ACT 2601, Australia.

(D) Corresponding author. Email: m.j.cracknell@utas.edu.au

Table 1. Summary of generalised classes for categorical input
variables: lithological units (GEO), soil classes (SOIL), vegetation
type (VEG) and land use (LAND)

ID    GEO                              SOIL

1     (Siluro-Devonian rift basin      Ca: leached sands (Ca6)
        sediments) argillaceous
        detrital sediment: shale,
        mudstone and siltstone with
        minor limestone and
        sandstone
2     (Permo-Triassic Sydney           Fa: vesicular loamy soils
        Basin) feldspar-or lithic-       and loamy soils having an A2
        rich arenite to rudite:          horizon (Fa2, Fa3)
        sandstone, conglomerate and
        quartzite
3     (Ordovician) high grade          Gb: deep porous, loamy soils
        metamorphic rock: migmatite      (Gb2, Gb4)
        and amphibolite
4     (Siluro/Devonian) igneous        Gd: dark shallow porous
        felsic/intermediate              loamy soils (Gd3)
        volcanic: andesite/rhyolite
        lava and tuff
5     (Predominantly Siluro-           Kb: dark cracking clays in
        Devonian) igneous felsic         association with brown
        intrusive: granite,              friable earths (Kb1)
        monzogranite, tonalite and
        granodiorite
6     (Siluro-Devonian) igneous        Ke: cracking grey clays with
        felsic volcanic: dacite,         small areas of cracking
        rhyolite and porphyry lava,      brown clays (Ke8)
        ignimbrite and tuff
7     Igneous foid-bearing             KK: organic loamy soils (KKI)
        volcanic: olivine analcime
        dolerite (associated with
        Cenozoic basalts)
8     (Predominantly Jurassic-         LA: water (LAKE)
        Cretaceous) igneous
        intermediate intrusive:
        syenite, mononite and quartz
        diorite
9     (Predominantly Permian)          LL: loamy soils with a A2
        igneous intermediate             horizon-yellow-brown earths
        volcanic: latite and             with stones and rock
        trachyte                         outcrops (LL1)
10    (Predominantly                   Mb: acid yellow leached
        Ordovician-Devonian) igneous     earths containing ironstone
        mafic intrusive: dolerite        gravel (Mb2, Mb5)
        and gabbro
11    Igneous mafic volcanic:          Md: red/brown friable earths
        basalt lava and tuff             with friable neutral red
        (Cenozoic basalts)               soils associated with
                                         cracking clays (Md3)
12    (Ordovician metasedimentary      Me: brown and red friable
        Adaminaby Group) marine          earths (Mel, Me2, Me3)
        sedimentary siliciclastic:
        turbiditic sandstone,
        mudstone and shale
13    (Silurian) meta-igneous          Mg: red friable porous
        mafic intrusive: amphibolite     earths (Mg 19, Mg9)
        and gabbro
14    (Silurian) meta-igneous          Mh: brown and red friable
        mafic volcanic: spilite          porous earths (Mh5)
15    (Late Cambrian-Devonian)         Mp: red friable earths (Mp1,
        meta-sedimentary                 Mp2)
        siliciclastic: slate,
        pelite, psammite, phyllite
        and schist
16    (Siluro-Devonian rift basin      Mr: acid leached yellow
        and Permo-Triassic Sydney        earths (Mr1, Mr2, Mr3, Mr4)
        Basin) quartz-rich arenite
        to rudite: terrigenous to
        shallow marine shale,
        siltstone, sandstone,
        quartzite and conglomerate
17    (Quaternary) regolith:           Mu: neutral leached red
        channel/floodplain alluvium,     earths (Mu4, Mu5, Mu6, Mu7)
        colluvium, lacustrine/swamp
        sediments, (coastal) sand
        dunes and plains
18    (Siluro-Devonian)                Mw: acid leached red and
        sedimentary carbonate:           yellow earths (Mw13, Mw14,
        limestone                        Mw15, Mw19, Mw7, Mw8)
19    (Late Cambrian-Ordovician)       NZ: sandy acidic gley soils
        sedimentary non-carbonate        and leached sands (NZ1)
        chemical or biochemical:
        chert
20    (Siluro-Devonian rift basin,     Pb: hard acidic red soils
        Permo-Triassic Sydney Basin      (Pb10, Pb11, Pb13, Pb5,
        and Cenozoic) sedimentary        Pb8, Pb9)
        siliciclastic: shale,
        mudstone, siltstone.
        sandstone, quartzite and
        conglomerate
21                                     Pd: hard acidic and neutral
                                         red soils (Pd2, Pd3)
22                                     Qb: hard neutral red soils
                                         (Qb11, Qb16, Qb8)
23                                     Qd: hard neutral/alkaline
                                         red/red mottled soils (Qd4)
24                                     Qr: friable neutral red
                                         soils (Qr2)
25                                     Rh: friable neutral brown
                                         soils with cracking clays
                                         (Rh2)
26                                     Ta: hard acidic yellow/
                                         yellow mottled soils (Ta6)
27                                     Tb: hard/neutral acidic
                                         yellow/yellow mottled soils
                                         (Tb 18. Tb22, Tb23, Tb24,
                                         Tb25, Tb26, Tb27, Tb28,
                                         Tb31, Tb33, Tb38, Tb39)
28                                     Ub: hard neutral yellow/
                                         yellow mottled soils (Ub27,
                                         Ub31, Ub32, Ub33, Ub34,
                                         Ub35, Ub36, Ub37, Ub39,
                                         Ub40)
29                                     Va: hard neutral/alkaline
                                         yellow mottled soils (Va21,
                                         Va22, Va23)
30                                     Wd: sandy acidic yellow
                                         mottled soils and/or leached
                                         sands (Wd5)

ID    VEG               LAND

1     Alpine complex    Conservation Area: national park,
                          state forest, tree lot and fenced
                          managed land
2     Cleared           Cropping: continuous or rotation
                          cropping and fodder crop
                          agriculture
3     Dry sclerophyll   Grazing: pasture and degraded land
        (shrub/grass)     (salt site, eroded area).
4     Dry sclerophyll   Horticulture: vegetable plot, orchard
        (shrub)           and vineyard (can be irrigated)
5     Forested          Intensive Animal Production--
        wetlands          piggery, dairy shed and horse
                          paddock
6     Freshwater        Mining and Quarrying: derelict,
        wetlands          operating, rehabilitated mine/
                          quarry
7     Grasslands        OUTSIDE: undifferentiated (ACT)
8     Grassy            Power Generation: energy corridor
        woodlands
9     Heathlands        River and Drainage System: river,
                          creek, dam, lake, lagoon and
                          estuary
10    Rainforests       Special Category: cliff, beach,
                          vegetated coastal dune
11    Saline            Transport and Other Corridors: road.
        wetlands          railway and airstrip
12    Wet sclerophyll   Tree and Shrub: native forest,
        (grassy)          softwood/hardwood plantation
13    Wet sclerophyll   Urban: residential, industrial and
        (shrub)           commercial land
14                      Wetland: swamp, marsh and mudflat
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

Table 2. Numeric variable Self-Organising Map-Hydrogeological
Landscape (SOM-HGL) attribute class label ranges based on 20th
percentile intervals for SOM node code-vector values (see Fig. 4)

DEM, Digital elevation model; TWI, topographic wetness index

Variable             Very low           Low          Moderate

DEM (m)            48.62-453.24    453.24-627.46   627.46-730.81
SLOPE (degrees)      1.49-5.01       5.01-6.95       6.95-9.42
TWI (ID)             6.35-7.17       7.17-7.45       7.45-7.8
RAIN (mm/year)     514.66-695.17   695.17-754.17   754.17-862.69

Variable               High          Very high

DEM (m)            730.81-899.46   899.46-1570.22
SLOPE (degrees)     9.42-13.52       13.52-22.2
TWI (ID)             7.8-8.17        8.17-9.66
RAIN (mm/year)     862.69-974.65   974.65-1567.49

Table 3. Hydrogeological Landscape (HGL) unit percent cover for
intersecting Self-Organising Map (SOM)-HGL units

Minimum and maximum percent are for a single SOM-HGL unit. Entries are
sorted in decreasing order of sum of top four SOM-HGL percent cover

HGL                 SOM-HGL        SOM-HGL          SOM-HGL
                      (n)            (%)             top 4%

                              Minimum    Maximum

Brothers               2         5.05      94.95     100.00
Kowmung                1       100.00     100.00     100.00
Lake George            6         0.07      98.43      99.86
Tinderry               5         0.72      58.27      99.28
Taralga               10         0.04      80.63      99.17
Monaro                10         0.05      82.62      99.03
Illogen Park          10         0.11      50.51      98.64
Palerang               8         0.38      62.26      96.98
Braidwood             14         0.07      84.37      96.76
Nadgigomar            10         0.07      50.84      96.43
Murrumbucca           16         0.08      81.51      95.85
Tabourie              21         0.04      54.61      95.77
Alluvial              19         0.03      89.69      95.56
Pejar                 10         0.10      55.16      95.18
Bombala               11         0.15      79.79      92.46
Kiandra               13         0.19      83.27      91.17
Bigga                 15         0.03      57.30      89.63
Tombong               16         0.07      61.56      87.97
Brown Mountain        19         0.19      78.55      87.70
Hawkins               26         0.01      54.24      86.96
Clear Range           14         0.09      31.67      85.83
Gerringong            25         0.07      47.84      85.82
Wagonga               13         0.34      59.90      85.54
Dalgety               17         0.03      68.28      84.60
Milton                15         0.39      64.23      83.46
Tantangara            19         0.04      43.12      82.87
Robertson             34         0.12      50.52      82.27
Long Flat             19         0.02      31.11      82.02
Shoalhaven Gorge      40         0.06      54.34      81.11
Shannons Flat         21         0.11      60.01      80.30
Canberra              23         0.02      39.16      80.13
Budderoo              33         0.01      43.39      78.50
Goobarragandra        14         0.05      33.47      77.60
Moruya                24         0.11      43.69      77.50
Lockyersleigh         15         0.07      37.25      76.14
Tarlo-Rhyanna         17         0.06      33.35      75.75
Windellama            20         0.10      35.31      75.39
Urialla               17         0.16      40.75      74.96
Nungatta              26         0.02      26.63      74.73
Pomeroy               28         0.02      48.25      74.57
Mogo                  22         0.01      27.26      73.99
Bungonia              24         0.03      35.07      73.90
Green Cape            25         0.08      49.43      73.45
Jerrawa               24         0.05      42.05      73.10
Adjungbilly           21         0.06      33.77      72.67
Nadgee                29         0.03      27.71      71.73
Black Range           31         0.02      24.33      70.70
Celeys                16         0.08      34.30      70.15
Woodhouselee          27         0.09      37.47      69.82
Deua                  42         0.01      25.64      69.47
Mongarlowe            20         0.04      21.35      66.88
Jinden                24         0.07      23.76      66.51
Quaama                31         0.01      46.81      65.54
Kybeyan               20         0.03      22.24      63.04
Chakola               34         0.02      22.67      62.86
Hollow Wood           24         0.16      22.95      62.46
Bulli                 21         0.07      25.00      62.23
Snowy                 22         0.04      31.73      62.11
Minuma                19         0.14      22.79      57.73
Hill Top              35         0.01      18.08      56.09
Cambewarra            29         0.03      23.56      56.09
Michelago             40         0.01      18.88      54.07
Araluen               24         0.04      17.93      50.00
Barneys Range         40         0.01      18.75      49.43
Greenwich Park        26         0.11      17.49      48.84

Table 4. Summary attributes for the top four (based on percent cover)
Self-Organising Map-Hydrogeological Landscape (SOM-HGL) units
intersecting Monaro (Fig. 8) and Windellama (Fig. 9) HGL units

DEM, Digital elevation model; TWI, topographic index

                                    Monaro
HGL:
SOM-HGL:          38            8            72           32

Cover

%                82.62        15.35         0.71         0.36
Cum. %           82.62        97.97        98.68         99.03
Pixels           9228          582          8419          440
Nodes              5            2            4             2

DEM

1[degrees]      V. high        High         High        V. low
%                0.57          0.66         0.43         0.72
2[degrees]       High        V. high      Moderate      V. high
%                0.37          0.32         0.2          0.16
Entropy          0.55          0.43         0.88         0.48

SLOPE

1[degrees]      V. low        V. low       V. low       V. low
%                0.64          0.63         0.58         0.59
2[degrees]        Low          Low          Low           Low
%                 0.2          0.23         0.21         0.23
Entropy          0.63          0.65         0.71         0.69

TWI

1[degrees]      V. high      V. high      V. high       V. high
%                0.57          0.59         0.48         0.40
2[degrees]       High          High         High         High
%                0.22          0.18         0.21         0.21
Entropy          0.73          0.73         0.85         0.92

RAIN

1[degrees]      V. low        V. low       V. low       V. high
%                0.68          0.95         0.46         0.49
2[degrees]        Low        Moderate       Low          High
%                0.14          0.04         0.28         0.25
Entropy          0.61          0.15         0.77         0.83

GEO

1[degrees]     Ig. mafic    Ig. mafic       sed.       Ig. inter
                 volc.        volc.        silic.        int.
%                0.82          0.79         0.85         0.74
2[degrees]    Marine sed.   Ig. felsic    Regolith    Sed. silic.
                silic.         int.
%                0.05          0.08         0.03         0.14
Entropy          0.26          0.27         0.22         0.29

SOIL

1[degrees]        Md            Md           Ub           Me
%                0.53          0.74         0.30         0.63
2[degrees]        LL            LL           Mr           Mu
%                0.15          0.26         0.22         0.15
Entropy          0.41          0.17         0.58         0.37

VEG

1[degrees]      Cleared       Grass.      Cleared       Cleared

%                0.90          0.64         0.85         0.83
2[degrees]       Grass       Cleared     Dry scler.   Wet scler.
                                          (shrub)      (Grassy)
%                0.04          0.32         0.08         0.04
Entropy          0.16          0.32         0.22         0.30

LAND

1[degrees]      Grazing      Grazing      Grazing       Grazing

%                0.93          0.76         0.88         0.86
2[degrees]     Tree and      Tree and     Tree and     Tree and
                 shrub        shrub        shrub         shrub
%                0.03          0.18         0.08         0.08
Entropy          0.14          0.28         0.19         0.22

                                  Windellama
HGL:
SOM-HGL:          16            18            37            24

Cover

%                35.31         19.59         11.32         9.17
Cum. %           35.31         54.90         66.22         75.39
Pixels           2485          8216           533          3085
Nodes              2             5             2             3

DEM

1[degrees]        Low          High          High          High
%                0.39          0.38          0.37          0.66
2[degrees]     Moderate       V. high      Moderate      Moderate
%                0.29          0.34          0.37          0.16
Entropy          0.80          0.84          0.76          0.64

SLOPE

1[degrees]      V. low        V. low        V. low        V. low
%                0.45           0.6           0.4          0.67
2[degrees]        Low           Low           Low           Low
%                0.24          0.22          0.20          0.21
Entropy          0.87          0.69          0.93          0.59

TWI

1[degrees]      V. high       V. high       V. low        V. high
%                0.24          0.46          0.27          0.46
2[degrees]       High          High         V. high        High
%                0.23          0.25          0.24          0.26
Entropy          0.99          0.85          0.98          0.82

RAIN

1[degrees]      V. low        V. low        V. low        V. low
%                0.40          0.44          0.56          0.50
2[degrees]     Moderate         Low           Low        Moderate
%                0.23          0.31          0.37          0.24
Entropy          0.88          0.76          0.54          0.76

GEO

1[degrees]    Marine sed.   Marine sed.   Marine sed.   Marine sed.
                silic.        silic.        silic.        silic.
%                0.92          0.88          0.90          0.94
2[degrees]    Sed. silic.    Ig. mafic    Ig. mafic.    Sed. silic.
                               volc.         volc
%                0.03          0.04          0.03          0.02
Entropy          0.13          0.20          0.16          0.10

SOIL

1[degrees]        Tb            LL            Tb            Ub
%                0.97          0.28          0.41          0.53
2[degrees]        Me            Tb            Ub            Pb
%                0.01          0.21          0.29          0.25
Entropy          0.02          0.62          0.45          0.42

VEG

1[degrees]    Dry scier,      Cleared     Dry scler,      Cleared
                (shrub)                     (shrub)
%                0.76          0.81          0.70          0.87
2[degrees]      Cleared     Dry scler.      Cleared     Dry scler.
                              (shrub)                     (shrub)
%                0.12          0.10          0.22          0.08
Entropy          0.35          0.27          0.34          0.19

LAND

1[degrees]     Tree and       Grazing       Grazing      Tree and
                 shrub                                     shrub
%                0.43          0.87          0.77          0.47
2[degrees]     Con. area     Tree and      Tree and        Urban
                               shrub         shrub
%                0.32          0.09          0.18          0.22
Entropy          0.48          0.20          0.28          0.55
COPYRIGHT 2016 CSIRO Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Cracknell, M.J.; Cowood, A.L.
Publication:Soil Research
Article Type:Report
Date:May 1, 2016
Words:9800
Previous Article:Soil microbial biomass carbon and phosphorus as affected by frequent drying-rewetting.
Next Article:Global warming as affected by incorporation of variably aged biomass of hairy vetch for rice cultivation.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters