Vianney Costemalle and Vincent Loonis (INSEE)
We all may be familiar with what is a city but when it comes to have a precise definition in order to produce statistics there is no consensus. Some typologies already exist such as the Tercet typology developed by Eurostat that aims at harmonising urban statistics accross European Union members. United Nations privilege the built-up area approach to define the city boundary and consequently the city size in its World Urbanization Prospects. For its national statistical purposes, France has for 60 years its own definition of a city, named urban unit, mixing built-up area and population threshold. We propose to compare 5 different methodologies to establish and characterize the agglomerations of population in France. The sources we rely on are the exhaustive geolocated databases for population and buildings, and open satellite data from the Copernicus program. The first methodology is the one traditionnaly used by Insee called urban unit. It consists in delineating contiguous urban area with at least 2000 inhabitants. The second method is the one suggested by UN for the sustainable developement goal indicator 11.7.1. It relies on satellite imagery to assess contiguous urban areas. The third method is the one provided by Eurostat to define the Cities, based on a 1km² population grid. We designed the fourth one based on this latter by rotating the 1km² grid a large amount of time and thus getting rid of any particular privileged direction. The last method is based on the famous clustering algorithm DbScan.