Objectives: Suburb boundaries were needed to assess the total suburb area. The objective of this research was to develop a novel geographic information system (GIS) application to delineate suburbs with boundaries corresponding to information contained in another data base comprising individual property records. The suburb boundaries derived in this manner may not relate to municipal boundaries, or sociopolitical boundaries, nor do they have to. The fundamentally correct suburb boundary would be the one encompassing what is perceived to be the suburb based on the suburb name in a particular data base that also contains other interesting attributes, such as water use, of individual properties.

Method: The ArcGIS environment was used to delineate suburbs by means of triangulated irregular network (TIN) modelling. Boundaries for suburbs with predominantly residential land use were created that included all residential properties according to the suburb name field as recorded in the treasury system. Other vacant areas were also included so as to obtain the total suburb area. The methodology was developed to assist research in the field of potable water services, but the method presented could be applied to other services that require management of information at suburb level.

Results: This article illustrates how a tedious task of suburb delineation could be automated in the GIS environment. The tool prevents subjective results that would be prone to error. The automated procedure described could effectively delineate a large number of predominantly residential suburbs in a relatively short time span and produce repeatable results. A reasonable outline could only be obtained if a sufficient number of parcels in the area contained the same suburb name. Functionality was added to the tool so that a limit could be set for this purpose. The default was that if more than 20% of the records were erroneous it was considered impractical to delineate a suburb. The derived suburb boundaries correspond to useful information in other data bases and would thus enable more effective management of the information.

Conclusion: A novel procedure to delineate suburb boundaries in the GIS environment was illustrated in this article. Information at two different spatial scales, namely, individual consumers and suburbs, could be married for the purpose of further research into suburban attributes. The tool was applied as part of a parallel research project to delineate 468 suburbs in this manner, results of which were submitted for publication elsewhere.

The application of geographic information system (GIS) tools in engineering and research into their effective application is not new. Research has been presented of GIS application in various engineering disciplines, such as public transport management (Dondo & Rivett 2004), sewer system analysis (Sinske & Zietsman 2002), river flood plain modelling (Yang, Townsend & Daneshfar 2006) and water master planning (Vorster et al. 1995).

Various methods are available for estimating residential water demand in South Africa, with a comprehensive review provided by Jacobs (2008). The most recent publications in this regard were by Van Zyl, Ilemobade & Van Zyl (2008) and Jacobs, Geustyn, Loubser and Van Der Merwe (2004). Further discussion of these water demand estimation methods is beyond the scope of this text; suffice it to say that all the available local methods for estimating water demand are based on the size of individual residential plots. Households living on larger plots use more water per day than those on smaller plots.

It may not seem clear at first why it would be important to create suburb boundaries as presented in this article. Formerly created suburb boundaries in the required format would certainly be available. The problem is that this may not necessarily be true for each suburb and, even if it were, the information does not necessarily link up between different data sets as desired for a particular research project. For example, predefined suburb boundaries were found to be dissociated in some instances from suburb names for individual plots in the treasury data base.

As part of a parallel research project into the development of a robust guideline for suburb-based water demand analyses (Jacobs, Sinske & Scheepers in press) it was necessary to evaluate a large number of suburbs in terms of various attributes, one of which was the total suburb area. An automated method was needed to delineate suburbs in order to obtain the total suburb area. The derived suburb boundaries needed to correspond to the available water use information for individual consumers stored in the treasury data base. The suburb boundaries derived in this manner may not relate to municipal boundaries or sociopolitical boundaries, nor do they have to. The fundamentally ‘correct suburb boundary’ would be the one encompassing all properties with the suburb name in a particular data base. Such a boundary may not exist nor may available boundaries be associated with the database to be analysed; it thus needs to be created.

It is apparent that GIS has found wide application in the field of engineering services, including water and planning. None of the applications addressed the need to match information at different spatial scales and delineate suburbs or any other similar area described by polygons as described in this text.

• The widely used ArcGIS Desktop 10.0 software package (licence type ArcInfo) from ESRI was used as GIS platform. ArcMap, which is the central application of ArcGIS, was used as the main spatial viewer. File management tasks were performed with ArcCatalog and spatial analyses with ArcToolbox, both part of the ArcGIS Desktop and accessed via ArcMap (ESRI 2010).

• ArcScene is a 3D visualisation application and part of ArcGIS, and was used in conjunction with the 3D Analyst tools of ArcToolbox to perform complex surface modelling for the suburb delineation process. The 3D Analyst extension of ArcGIS is a system requirement for the above.

• The end results were finalised via ModelBuilder, which is part of ArcGIS.

Chang (2010) defines a feature as any representation of a real-world object on a GIS-map; it could be any shape. In ArcGIS, a feature class stores spatial features of the same geometric type (i.e. point, line, polygon, etc.), same attributes i.e. common set of attributes) and the same spatial reference (i.e. common mapping co-ordinate system). In the above context, for example, all the parcels addressed by the delineation procedure would be stored in a feature class. Feature classes again are stored in an ArcGIS geodatabase as either standalone or grouped in a feature dataset. These terms are applicable to this study as defined below.

The treasury data base would contain information for each of these 500 consumers or residences. Each would have a water meter read monthly, with data stored in the treasury data base. Each consumer’s property would be described by numerous fields in the data base. One of these fields would be the suburb name field with the entry: Suburb A. The town planner would be able to provide an independent GIS data base describing the cadastral layout of the town - this could be seen as a map of the town showing all the properties. This GIS data base typically contains the suburb name and land use for each property in separate data fields. The suburb name in the GIS data base would not always match up with the suburb name in the treasury data base.

In between these 500 parcels comprising residential plots would also be vacant areas that would typically represent roads, parks or public open spaces, but these would not typically be flagged with the suburb name. These vacant areas are often not specifically captured as parcel polygons. It would be obvious to the reader that all the parcels, roads and other vacant areas in between the parcels should actually be part of Suburb A if the total suburb area needs to be considered.

Readily available up-to-date polygons (in GIS format) depicting suburbs in the desired fashion are unfortunately seldom available. Boundary and name changes over time, particularly after local political change in the mid–1990s, resulted in lacunae. This was true for boundaries at the provincial level to suburb and ward level. Another problem was that of duplicate suburb names, for example, a study that would encompass the entire country and where the same suburb name would be found in different cities. The suburb boundary matching desired attributes and encompassing all spaces in between plots could easily and quickly be generated by the method reported in this study, producing repeatable results.

Different methods of interpolation are available to form these triangles. The most widely used is called Delaunay triangulation and is implemented in the ArcGIS software suite (ESRI 2010). This triangulation method ensures that all sample points are connected with their two nearest neighbours to form triangles as equiangular or compact as possible. In this manner, it is possible to avoid the formation of too many unwanted sharp, long and skinny triangles (ESRI 2010; Chang 2010; Li & Ai 2010). A finished TIN comprises three types of geometric objects, namely, (1) triangles (facets), (2) points (nodes) and (3) lines (edges). Elevation data is stored at the nodes, whereas slope and aspect data are stored for each facet and remain constant over the facet (Chang 2010). Most GIS software packages implement TIN because one of their data structures and have the ability to export the abovementioned individual components of the TIN as separate polygon, point and line features for further analysis.

Apart from TIN, most GIS also implement the grid structure for terrain modelling. One of the biggest advantages of the TIN model over the grid model is the flexibility of TIN to model more detail at certain locations (i.e. terrain specific source data such as roads, rivers, lakes and parcels can be incorporated in the triangulation process). Only high-resolution grids can show these detailed features, but this would not be an optimal solution with regard to data storage because the cell size (which is constant for the grid) will have to be defined as very small over the whole study area (Burrough 1986; Burrough & McDonnell 1998; Chang 2010).

A TIN data model can also be used to represent and model two dimensional (2D) surfaces, as is the case with this research. Li and Ai (2010) discussed the application of Delaunay TIN to detect various spatial and structural characteristics hidden in 2D geometry data (i.e. a type of spatial data mining application). They also pointed out an important aspect of Delaunay TIN, namely, that the triangle element can play two roles: either the component of the polygon feature or the bridge between neighbouring objects. Triangles playing the bridging role are distributed on the principle of ‘nearest connection’ of Delaunay TIN. Hereby, the neighbourhood relationship is presented by only one triangle no matter how far between these objects, which is a useful characteristic for spatial neighbourhood analysis.

The suburb delineation process presented in this article is also a 2D TIN application based on the abovementioned dual role of the Delaunay TIN triangle. This means that some triangles will be used to cover the entire area of parcels in the suburbs (i.e. TIN triangles located on parcels) and others will span the empty space between parcels (i.e. a TIN triangle located in the empty space bridging the gap between two other TIN triangles located on parcels). The method can delineate a large amount of suburb boundaries all at once and can be executed in ArcGIS via ten geoprocessing steps (Figure 1).

FIGURE 1: Suburb delineation procedure.

The elevation field (in the Parcels feature class) can be filled with any constant elevation value (such as e.g. 1 m) for all parcels (Step 2 of Figure 1) because the TIN will be used for 2D analysis only. The suburb code in the Parcels feature class (Step 2 of Figure 1) must be a unique integer code for each suburb name. The TIN model can best be created in the ArcScene environment with the Create TIN tool (Step 3 of Figure 1), accessible from the integrated ArcToolbox (note, the 3D Analyst extension of ArcGIS is required). The 2D TIN model will be built based on the parcel vertices, which all have the abovementioned 1 m spot height allocated. Important is to set the surface type to Softvaluefill. This will ensure that the parcel boundaries will be enforced in the triangulation as breaklines (i.e. TIN triangle edges will not cross the parcel boundaries [Figure 2a]). Furthermore, the TIN triangles inside these parcel polygons will hereby be attributed (i.e. filled) with the corresponding suburb code tag value (for cross-reference checking). These TIN triangles can now be extracted from the TIN model and saved as a new polygon feature class (Step 4 of Figure 1) via the ArcToolbox conversion tool TIN Triangle.

FIGURE 2: Transforming (a) individual parcels to (b) suburb areas using a triangulated irregular network (TIN).

FIGURE 3: Land use summarisation.

The final selection of TIN triangles can now be dissolved (Step 10 of Figure 1) based on the suburb names from the above spatial join results in order to obtain the final delineated suburb polygons. It can be recalled that this boundary does not need to match up to any other boundary - it needs to delineate the outer edge of a number of plots that were flagged in a given data base as being part of this suburb, plus all spaces in between.

Composite land use information needs to be summarised per suburb (Step 1 inFigure 3) by generating a summary table on the Parcels feature class. This can be accomplished in ArcMap with the integrated ArcToolbox function Summary Statistics and specify the Suburb and Land_use fields as the two Case fields on which the summary should be based. The resultant summary would contain the various information required as output in multiple suburb records and could be linked with the Suburbs feature class (containing the final delineated suburb boundaries) via a one-to-many relationship. This type of relationship would, however, make further processing by the model unnecessarily complex.

A simpler link could be accomplished in ArcMap by selecting and exporting from the composite table the abovementioned land uses into separate tables and consecutively joining them (i.e. one-to-one relationship) with the Suburbs feature class (Step 2 and Step 3 of Figure 3). The Suburbs feature class now contains one record per suburb with the land use information contained in separate fields, which is ideal for further processing. The procedure checks that the number of parcels per suburb deviate less than 20% from the number of parcels as recorded for the corresponding suburb in the treasury database. Suburbs with more than 20% deviation in number of parcels are excluded from the selection because in these cases there are obvious fundamental differences in the suburb boundaries between the two data sets and the delineated suburb would not be considered valid for the purpose of deriving its total area.

• The semi-automatic suburb delineation process and the land use summarisation process could be implemented directly as models in the ModelBuilder environment, subsequently reducing the analysis time.

• The invalid (long and skinny) TIN triangles located on the edges of the suburb after the TIN interpolation process add inaccuracies to the suburb delineation procedure in some cases. An improved selection of TIN triangles could be obtained with almost no invalid triangles on the suburb edges by reducing the threshold settings. Only a few valid triangles located on wide roads and traffic circles would be wrongly missed by this finer threshold. The distance threshold cannot be reduced significantly for study areas containing suburbs with many unidentified (vacant) land areas (i.e. those not captured as parcels) because these vacant areas would then wrongly be excluded from the suburb area calculation.

Burrough, P.A., 1986, Principles of geographical information systems for land resources assessment, Oxford University Press, Oxford.

Burrough, P.A. & McDonnell, R.A., 1998, Principles of geographical information systems, Oxford University Press, Oxford.

Chang, K.T., 2010, Introduction to geographic information systems, 5th edn., McGraw-Hill, New York.

Coetzee, S. & Bishop, J., 2009, ‘Address databases for national SDI: Comparing the novel data grid approach to data harvesting and federated databases’, International Journal of Geographic Information Science 23(9), 1179–1209. http://dx.doi.org/ 10.1080/13658810802084806

Coetzee, S. & Rademeyer, I.M., 2009, ‘Testing the spatial adjacency match of the Intiendo address matching tool for geocoding of addresses with misleading suburb or place names’, In-proceedings International Cartography Conference, Santiago, Chile, November 15–21, 2009, n.p.

Dondo, C. & Rivett, U., 2004, ‘Spatial information systems in managing public transport information’, South African Journal of Information Management 6(2), 8 pages.

ESRI, 2010, ArcGIS Desktop Help, Environmental Systems Research Institute, Redlands, CA.

Jacobs, H.E., Sinske, S.A & Scheepers, H.M., in press, article submitted for peer review and possible publication to Urban Water Journal, submitted September 2012.

Jacobs, H.E., 2008, ‘Chronologiese oorsig van Suid-Afrikaanse riglyne vir residensiële gemiddelde jaarlikse waterverbruik met erfgrootte as onafhanklike veranderlike: Navorsings- en oorsigartikel’, Suid-Afrikaanse Tydskrif vir Natuurwetenskap en Tegnologie 27(4), 240–265.

Jacobs, H.E., 2008, ‘Residential water information management’, South African Journal of Information Management 10(3), 12 pages.

Jacobs, H.E. & Fair, K.A., 2012, ‘A tool to increase information-processing capacity for consumer water meter data’, South African Journal of Information Management 14(1), 7 pages.

Jacobs, H.E., Geustyn, L.C., Loubser, B.F. & Van Der Merwe, B., 2004, ‘Estimating residential water demand in southern Africa’, Journal of South African Institution of Civil Engineering 46(4), 2–13.

Li, J. & Ai, T., 2010, ‘A triangulated spatial model for detection of spatial characteristics of GIS data’, in Y. Wang (ed.), Proceedings of the 2010 IEEE International Conference on Progress in Informatics and Computing, PIC 2010, Shanghai, China, 10–12 December 2010, IEEE, Washington, DC, pp. 155–159.

Maguire, D.J., 1992, An overview and definition of GIS, in D.J. Maguire, M.F. Goodchild, & D.W. Rhind (eds.), Geographical information systems: Principles and applications, Vol.1, pp. 9–20, Longman, New York.

Obermeyer, N.J. & Pinto, J.K., 1994, Managing geographic information systems, The Guilford Press, New York.

Shekhar, S. & Chawla, S., 2003, Spatial databases: A tour, Pearson Education, Inc., Upper Saddle River, NJ.

Sinske, S.A. & Zietsman, H.L., 2002, ‘Sewer-system analysis with the aid of a geographical information system’, Water SA 28(3), 243–248. http://dx.doi.org/10.4314/wsa.v28i3.4891

Sinske, S.A. & Zietsman, H.L., 2004, ‘A spatial decision support system for pipe-break susceptibility analysis of municipal water distribution systems’, Water SA 30(1), 71–79. http://dx.doi.org/10.4314/wsa.v30i1.5029

Van Zyl, H.J., Ilemobade, A.A., & Van Zyl, J.E., 2008, ‘An improved area-based guideline for domestic water demand estimation in South Africa’, Water SA 34(3), 381–392.

Vorster, J., Geustyn, L.C., Loubser, B.F., Tanner, A. & Wall, K., 1995, ‘A strategy and master plan for water supply, storage and distribution in the East Rand region’, Journal of South African Institution of Civil Engineering 37(2), 1–5.

Wikipedia, 2012a, ‘Greenfield land’, viewed 22 August 2012, from http://en.wikipedia.org/wiki/Greenfield_land

Wikipedia, 2012b, ‘GIS’, viewed 22 August 2012, from http://en.wikipedia.org/wiki/GIS

Yang, J., Townsend R.D. & Daneshfar, B., 2006, ‘Applying the HEC-RAS model and GIS techniques in river network floodplain delineation’, Canadian Journal of Civil Engineering 33(1), 19–28. http://dx.doi.org/10.1139/l05-102