Monitoring land use in cities using satellite imagery and deep learning

Over time, cities expand their physical footprint on land and new cities emerge. The shape of the built environment can affect several domains which are policy relevant, such as carbon emissions, housing affordability, infrastructure costs, and access to services. This study lays a methodological ba...

Full description

Saved in:
Bibliographic Details
Published inOECD Regional Development Papers
Main Authors Banquet, Alexandre, Delbouve, Paul, Daams, Michiel, Veneri, Paolo
Format Journal Article
LanguageEnglish
Published Paris Organisation for Economic Cooperation and Development (OECD) 13.06.2022
Subjects
Accuracy
Area planning & development
Cities
Classification
Commuting
Deep learning
Entrepreneurship
Land use
Metropolitan areas
Per capita
Regions
Online AccessGet full text

Cover

More Information
Summary:Over time, cities expand their physical footprint on land and new cities emerge. The shape of the built environment can affect several domains which are policy relevant, such as carbon emissions, housing affordability, infrastructure costs, and access to services. This study lays a methodological basis for the monitoring and consistent comparison of land use across OECD cities. An advanced form of deep learning, namely the U-Net model, is used to classify land cover and land use in EC-ESA satellite imagery for 2021. This complements conventional statistical data by monitoring large surfaces of land efficiently and in near real-time. In specific, following the availability of detailed data for model training, built-up areas in residential or business-related use are mapped and analysed for 687 European metropolitan areas, as a case application. Recent urban expansion’s speed and shape are explored, as well as the potential for assessing land use in cities beyond Europe.
ISSN:2709-4065
DOI:10.1787/dc8e85d5-en