Cities are often much larger social and physical systems than their governmental or municipal boundaries would suggest. Using unsupervised machine learning, high resolution satellite imagery, and geographic data science, this problem of urban boundary mismatch can be interrogated. Analysing cities in the US, Canada, the UK, Nigeria and South Africa, a globally-applicable method to analyse this boundary mismatch is possible.
Explaining the science
The project uses high resolution multi-band satellite imagery, hierarchical clustering, and computational geometry methods to construct 'apparent' urban areas. At its core, the satellite image is classified into built-up or not built-up space. Then, areas that have a similar, consistent pattern of built-up land are collected together as an urban area. Finally, the tightest containing shape is drawn around the land collected into the same urban area.
This project aims to build open source tools for planners and analysts to build 'apparent' urban areas for cities from satellite imagery, and identify how the political and government systems in these city areas are split by government boundaries.
The aim is to come up with a better and more globally-useful method to identify where 'the city' begins and ends, as well as measures of how an urban area is split by multiple governments. This is important work, since ill-aligned boundaries and ill-coordinated urban governments have serious costs on urban efficiency, health, economic performance and the behaviour of urbanites.
The work can benefit remote sensing and urban data science companies in industry, such as Mapbox, MapZen, CARTO, Geolytics and Enigma.