Contributions to the formalization and implementation of spatial urban indices using open data: application to urban sprawl studies

Mots clé :

• Urban sprawl
• Land use
• Open data
• Population density

Urban processes take place as a consequence of different interacting factors, linked between them in such a way that the resulting process is complex to measure and understand. Given the increasing number of people living in cities, understanding the underlying complexity of these urban patterns is thus becoming a pressing issue. To this end, we propose in this thesis decision-support tools applied in the context of urban analysis, which allow to study land use mix and urban sprawl phenomena.
In our first contribution, a framework for capturing spatial land use mix in cities is presented. In the first place, urban data are extracted from OpenStreetMap. Using Kernel Density Estimation techniques, land use density estimations are carried out for residential and activity uses. The outputs are employed to calculate spatial mixed-use development indices. Additionally, density estimations for different activity types (i.e. commercial and industrial, leisure and amenities, and shops) are proposed. We provide fine-grained Geographic Information System outputs, which happen to be an asset particularly for urban planners, supporting and aiding their decision-making procedure -- specially in relative comparison to aggregated measures.
In a second contribution, the above work was extended for calculating spatial urban sprawl indices. The proposed approach formalizes sprawl under a sustainable development angle, namely into three dimensions: land use mix, dispersion of built-up area, and accessibility to activity opportunities. This results in a manageable number of dimensions, where each dimension is formalized in an easy-to-interpret way, and in particular pertinence to the aspects of sprawl that impede sustainable development.
In our third contribution, we propose two approaches for performing disaggregated population estimates. The first one exploits information on residential surfaces, assuming a constant residential surface consumption per-capita. By means of employing gridded -- i.e. aggregated -- census tract data, a fine disaggregation is carried out to distribute population count data into buildings. The second one consists of a fully convolutional neural network that maps coarse-grained to fine-grained population data. We use the OpenStreetMap database to extract a set of urban features which describe a local urban context and guide the disaggregation procedure. Population densities are estimated for grid-cells 25 times smaller than the input resolution, i.e. 200m by 200m.

Directeurs:

• Mr Peter Sturm (Directeur de recherche - INRIA )
• Mr Serge Fenet (Maître de conférence - Université Lyon 1 )

Raporteurs:

• Mr Gilles Gesquiere (Professeur - Université Lumière Lyon 2 )
• Mr Jochen A.G. Jaeger (Associate Professor - Concordia University )

Examinateurs:

• Mme Silvia Ronchi (Ingenieur de recherche - Politecnico di Milano )
• Mr Nour-Eddin El Faouzi (directeur de recherche - IFSTTAR )
• Mr Marc Bourgeois (Maitre de conférence - Université Lyon 3 )