Visualizing Australia’s urban extent: a comparison between residential housing addresses and night-time light data

ABSTRACT This graphic uses the 2014 Geocoded National Address File (G-NAF), a geocoded address database containing the physical addresses of all properties in Australia, to extract residential property addresses, and then uses Baidu eCharts visualization library to map out the extent of urban areas in Australia and its major metropolitan regions. By comparing with the DMSP-OLS night-time light data we identify the advantages of using G-NAF for mapping urban extent and its potential in monitoring and modelling urban growth and change.

Detecting the extent of urban areas is a continuous effort by geographers and planners in order to plan, manage and optimize urban growth. A number of methods have been developed to identify the urban extent, including the classification of land uses from satellite imagery (Trianni et al., 2015), the population density approach (Linge, 1965) and the use of night-time light data (Henderson, Yeh, Gong, Elvidge, & Baugh, 2003;Imhoff, Lawrence, Stutzer, & Elvidge, 1997;Shi et al., 2014;Sutton, Goetz, Fildes, Forster, & Ghosh, 2010). While widely applied, these approaches have their inherent limitations, including classification and accuracy issues (Huang, Li, Liu, & Li, 2016), the varying population density standards adopted in different country contexts (Herbert & Thomas, 1997), and the inherent limitation with night-time light data (Henderson, Storeygard, & Weil, 2012).
We extracted all residential property addresses from the 2014 Geocoded National Address File (G-NAF) data (see https://www.psma.com.au/products/g-naf), which is a geocoded address database containing the addresses of all properties in Australia, and used ArcGIS 10.2 software to compute the number of properties at a spatial resolution of a 1 km 2 grid for the whole of Australia (PSMA Australia, 2018). Subsequently, we used the Baidu eCharts (Baidu eCharts; see http:// echarts.baidu.com/) visualization library to visualize the residential address data. Figure 1 illustrates all address points with at least five addresses in each grid, classified into three categories: scattered residential areas (with 5-15 addresses in each gird), rural residential areas (16-76 addresses) and urban residential areas (> 76 addresses). The distinction between urban and rural residential areas is based on the minimum standard of 200 persons/km 2 adopted by the Australian Bureau of Statistics (ABS) to delimitate urban areas and an average household size of 2.6 persons (ABS, 2015), resulting in approximately 77 properties/km 2 . Insert maps of the major metropolitan areas of Perth, Adelaide, Melbourne, Canberra, Sydney and Brisbane-Southeast Queensland are included in Figure 1.
Compared with the 2013 DMSP-OLS night-time light data (see https://ngdc.noaa.gov/ eog/dmsp/downloadV4composites.html) resampled to a 1 km 2 grid to match with the 2014 G-NAF data, it is evident that the settlement areas represented using the residential address data are more accurate than the night-time light data. In all metropolitan areas along the coasts, the extent of urban areas illustrated in Figure 2 extend to outside the coastal boundaries due to the intensity of light along the shoreline. In contrast, the G-NAF data represent only residential locations when delimitating the urban areas. Similarly, urban and rural residential areas in the inland region are also more accurate with the G-NAF data ( Figure 1) compared with the night-time light data, where the latter falsely depicted more 'urban areas' due to non-residential lights (such as street lights or bush fires). Nevertheless, it would be useful to use a combination of population census, residential property addresses as well as night-time light data to define the extent of urban areas more accurately.
Future research opportunities exist using the property address data for continuous monitoring of the change of the urban extent, identifying and modelling the driving factors to urban growth, as well as predicting future growth scenarios and directions.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.