The Centre for Australian Research into Access (CARA) has been utilising the fast-calculating methods of the Pandana Python library in combination with Geopandas methods to perform continental-scale network analysis. CARA's objective is to provide address-level accessibility calculations for health and education with a particular focus on rurality, to assist academic researchers and to inform policymakers. The work we are producing is a result of an Australian Research Council Linkage Infrastructure, Equipment and Facilities (LIEF) grant, which has recently finalised building a spatially detailed infrastructure for a more equitable nation, representing a digital twin for modelling the patterns and processes impacting the Australian population.

CARA has adapted open-source Python libraries, file formats, and data to develop network analysis methods that operate on continental scales that have previously been too costly to process. The Pandana open-source Python library enables fast network calculations to find shortest paths using contraction hierarchies (Foti & Waddell 2012). We have used methods from Pandana to perform ‘number of nearest’ calculations (n-nearest), matrix calculations, and catchment area calculations for both time and distance impedances. Calculations have been carried out using origin data for approximately 10.6 million residential addresses across Australia to various health and education destination datasets. Spatial methods from the Geopandas library were also used to account for the times and distances between origin/destination points and the network, and also to aggregate outputs to spatial units such as those in the Australian Statistical Geography Standard (ASGS).

Outputs from our n-nearest network calculations using the Pandana nearest_pois method (3.7 million nodes, 7.8 million edges, two impedances, and 7,073 destinations) have been completed in less than 3.5 minutes. Outputs from our matrix and catchment network calculations using the Pandana shortest_path_lengths method have fluctuating run times due to the varying number of input origin and destination points with each calculation. However, they are robust in handling a large number of inputs.

These methods have recently been adapted to use Overture Maps data that will allow them to be applied to a wide variety of countries throughout the world.