Spatial aggregation methods: an interactive visualization tool to compare and explore automatically generated urban perimeters

International audienceChoosing the appropriate scale of analysis is a well-known problem in regional studies. Changing the level of spatial analysis is not trivial and can have substantial effects on the resulting the indicators, their representation and their interpretation, especially if the original data is spatially diverse and at a fine granularity. One cannot regroup locational data as spatial data points automatically without consequences. As geographical datasets are becoming increasingly available, with a finer resolution, and as territorial / spatial clustering algorithms are becoming easier to apply on vast amounts of spatial data, we want to provide a geospatial application, which could help to present and analyse this issue. What definitions can be used to delineate the idea of a city, an urban area or an agglomeration? At what size, population density, volume of urban activity or surface are we observing a city? We consider this issue especially in a methodological comparative purpose taking the example of analysing the volume of scientific activity of European cities. In this proposal, we provide a comparative analysis of spatial clustering methods or aggregation procedures. In support of the presentation, we propose an interactive web-based application designed to explain and (geo)visualize their effects on different results: volume of urban activity (discrete values) and city rankings (ordinal values), on the one hand; and effects on the spatial configuration, on the other hand. In this exploratory work, we will focus on functional data about the geography of scientific production. To delineate automatically functional perimeters of European cities, we use the distribution of scientific activities i.e. the number of publications per municipalities computed from the Web of Science database following an extensive process of geo-localisation of authors' addresses (Eckert et al., 2013). We will compare the results of several spatial aggregation methods applied on these geolocalized points: Hierarchical Clustering using various aggregation functions and criteria with or without weighting, then density-based methods including DBSCAN and HDBSCAN. We will use the volume of scientific publications associated to each geolocalized points as a weighting indicator