*2.1. Study Area*

Darwin (12.4634◦ S, 130.8456◦ E) is situated on the north coast of Australia, a landscape dominated by tropical savanna (Figure 1).

The population of approximately 140,000 [32] constitutes an increase of nearly 60,000 people in twenty years [33]. The climate is monsoonal and experiences distinct annual dry (May to September) and wet (November to April) seasons with transitional periods in between. Mean annual rainfall is approximately 1700 mm; the mean minimum and maximum temperatures range from 19.3◦ to 25.3◦ and 30.6◦ to 33.3◦, respectively [34]. Compared to other major Australian capital cities, housing density is low; under 20 private dwellings per square kilometre as opposed to between 150 to 200 per square kilometre in Sydney and Melbourne [35]. A combination of urban and periurban environments in the Darwin region provides resources for avian populations that are typically unavailable outside of this area in the dry season.

**Figure 1.** Study area of the Greater Darwin region (map data sources: Esri, DeLorme, HERE, USGS, Intermap, iPC, NRCAN, Esri Japan, METI, Esri China (Hong Kong), Esri (Thailand), MapmyIndia, Tomtom).

### *2.2. Spatial Data*

To investigate broad-scale changes in urbanisation, Landsat satellite imagery of the study area from April 1998 and 2018 was obtained from the United States Geological Survey's Global Visualization Viewer (GloVis), with 1998 imagery obtained from the Landsat 4 and 5 Thematic Mapper (TM) satellite and 2018 imagery taken by the Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) (Tables A1 and A2). April marks the end of the wet season in Darwin, allowing for images with minimal cloud cover and maximum vegetation growth. Images were clipped in ArcGIS version 10.4.1 [36] to a shapefile of Darwin region localities provided by the Northern Territory Government Department of Environment and Natural Resources and then imported into SAGA GIS version 7.3.0 [37]. Classification of land use types was carried out via geographic objectbased image analysis (GEOBIA). In traditional pixel-based image classification, classes are assigned per pixel; however, GEOBIA uses segmentation and classification to better replicate what the human eye perceives [38,39]. The segmentation process combines pixels of similar spectral properties into objects in the form of polygons, and these can then be classified using either supervised or unsupervised techniques [40]. After automated segmentation in SAGA GIS, the two clipped images were classified by assigning 'training sites' (essentially selecting a minimum number of polygons and ascribing them a land use type) and then running a supervised classification. The resulting vector layers were then manually edited using the original satellite image to reassign any misclassified polygons. The initial uncorrected GEOBIA and user-corrected images were then re-imported back into ArcGIS, where fifty accuracy assessment points were randomly generated and an error matrix was constructed to assess both the producer (SAGA) and user (human) accuracy when assigning classification. For all map classifications satellite imagery, aerial photographs and Google Earth Pro version 7.3.2.5776 (64 bit) were used to assist in the

accuracy assessment; however, due to the retrospective nature of the earlier imagery, only the 2018 images could be further checked, if required, using ground control points.

#### *2.3. Bird Survey Data*

Survey data were extracted from the BirdLife Australia 'Atlas of Australian Birds' database (hereafter referred to as the 'Bird Atlas') for the years 1998 and 2018. Several types of surveys compiled these data: systematic bird surveys of 2 ha, 5 km and 500 m; unstandardised bird surveys, either along a fixed route or incidental; bird list surveys and the Shorebird 2020 surveys (a record of shorebird sightings not necessarily in coastal habitats). All records include a location, latitude and longitude, dates and species common names. In most records, a time is recorded and whether there is any breeding activity. Sighting notes of interest are sometimes included.

As the focus of the project was on terrestrial, predominantly diurnal species, Bird Atlas records were excluded if the species was almost exclusively nocturnal, was a waterbird or seabird (except Magpie Geese, *Anseranas semipalmata*), or the species was considered 'vagrant'. Using information from BirdLife Australia [41], the Atlas of Living Australia [42] and Australian Bird Data Version 1.0 [43], the feeding preferences of species were categorised from most preferred to occasional.

To give a general overview of any assemblage changes, the records of 1998 and 2018 were categorised into the following primary food sources: fruit, insects, invertebrates, nectar, omnivore, raptor, scavenger, seed, vegetation or vertebrate. If a species was considered to feed on two types of food source equally, both were considered the primary food source. Species were sorted by their level of specialisation: whether they had one, two, three or more food sources (Table A3).
