How does Geoscape ‘see’ surface cover?

Surface cover is one of three themes contained within Geoscape®. The other two are buildings and trees.

Geoscape combines advances in satellite imagery, machine learning and big data processing to create a digital representation of Australia’s built environment. It’s the first time location-based information and data analytics capability have been combined and made available in this way on a national level.

PSMA’s data products have traditionally used vector data – points, lines and polygons – to represent address points, roads and property boundaries etc.

Geoscape uses vector data to represent the location of buildings, but also makes use of raster data to represent variation in surface cover across Australia (and tree coverage across the built environment).

What is raster data?

Raster data is used to represent thematic or continuous information and is commonly derived from aerial photography or satellite imagery. It consists of a matrix (grid) of cells (pixels), with each cell assigned a value. The values represent the information being portrayed by the dataset. In the case of Geoscape, values represent real-world geospatial information including surface cover: bare earth = 2; road and path = 3; grass = 4; and so on.

Geoscape’s surface cover theme operates at two levels – a high resolution two metre dataset representing surface cover within urban areas and a national resolution 30 metre dataset representing surface cover nationally. This means a grid measuring two metres by two metres is applied to satellite imagery of urban areas and remote communities, and a grid of 30 metres by 30 metres is applied to cover the remainder of the Australian continent.

Conceptual representation of a raster layer [Satellite image used under licence from Spatial Services, NSW Government.]
Cell values for Geoscape’s surface cover theme at the two metre level

Geoscape is made possible by the constellation of satellites that are part of the DigitalGlobe platform. They deliver a suite of satellite images. Electromagnetic spectral analysis and classification processes then use algorithms to extract thematic values from the satellite-captured imagery.

The Electromagnetic Spectrum

Electromagnetic energy travels in waves. The electromagnetic spectrum covers the range, in wavelengths (and frequencies), of all electromagnetic energy.

Visible light is the part of the electromagnetic spectrum the human eye can detect. It falls within the spectrum at wavelengths between about 400 and 700 nanometres. Materials or surfaces absorb electromagnetic energy at certain wavelengths and reflect it at others. By differentiating these differences, the human eye sees colour. For example, grass reflects electromagnetic energy at wavelengths of about 510 nanometres. We interpret that as green.

The sun emits electromagnetic energy (solar radiation) largely as light and heat in the ultraviolet, visible light and infrared portions of the electromagnetic spectrum. The Earth’s surface absorbs and reflects this to varying degrees depending on the nature of the surface. Satellite sensors interpret reflections similarly to the human eye, but they are able to detect a greater portion of the electromagnetic spectrum. This is key to the analysis of satellite imagery used to build Geoscape.

Geoscape’s Surface Cover theme

By applying a grid over satellite imagery, using machine-learning to interpret what’s reflected in each cell based on known patterns of electromagnetic radiation reflectance, and then assigning an appropriate value to each cell, Geoscape represents surface cover. It sees bare earth, roads, grass, water, buildings and more. (More detail on Geoscape’s surface cover classification can be found in Table 43 on page 77 of the Geoscape Data Product Description.) Geoscape builds a pixel by pixel understanding of the surface of the Australian continent – a top-down look.

It’s a process not without its challenges. Features across the Australian landscape and built environment don’t fit neatly into any sort of grid. Challenges lie in interpreting the mix of the environment and the reflectance of surfaces across the electromagnetic spectrum for each of the 2 x 2 or 30 x 30 metre cells.

At PSMA, we take strips of data – raster layers – and mosaic them together to provide coverage of Australia’s 7.6 million square kilometres. At the edges of strips, or for multiple strips covering the same area collected at different times, we overlay the data layers and bring the best information to the top. This would be either the latest data capture or that where the impacts of cloud cover and shadow are minimised. We then clip the strip(s) to a defined location. The result is seamless coverage.

The rich information that’s gathered through this and other processes and contained within Geoscape is valuable to many. For example, planners benefit from having visibility of pervious and impervious surfaces to understand water absorption and run-off potential. By understanding the extent of impervious surfaces and a given volume of rainfall, planners can estimate the amount of water likely to build up within a catchment.

It’s an art, as much as it is a science, to classify the environment using remote sensing techniques. The Australian landscape and built environment doesn’t fit neatly into either vector or raster data types. Geoscape’s combined use of vector and raster data is likely to give rise to new data themes over time.

For further information

Further detail on the attributes contained within Geoscape can be found in the Geoscape Data Product Description and by viewing our on-demand Technical Architecture webinar.

A Geoscape Sample Dataset is also available for direct download. It covers an area of approximately 85 square kilometres that includes the South Australian township of Freeling and a surrounding rural area.

Geoscape provides unparalleled coverage of the Australian built environment, delivered to suit location-enabled analytics and 3D mapping tools.