EO + morphometrics : understanding cities through urban morphology at large scale

Wang, Jiong and Fleischmann, Martin and Venerandi, Alessandro and Romice, Ombretta and Kuffer, Monika and Porta, Sergio (2023) EO + morphometrics : understanding cities through urban morphology at large scale. Landscape and Urban Planning, 233. 104691. ISSN 0169-2046 (https://doi.org/10.1016/j.landurbplan.2023.104691)

[thumbnail of Wang-etal-LUP-2023-EO-morphometrics-understanding-cities-urban-morphology]
Preview
 Text. Filename: Wang_etal_LUP_2023_EO_morphometrics_understanding_cities_urban_morphology.pdf
Final Published Version
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (21MB)| Preview

Abstract

Earth Observation (EO)-based mapping of cities has great potential to detect patterns beyond the physical ones. However, EO combined with the surge of machine learning techniques to map non-physical, such as socioeconomic, aspects directly, goes to the expense of reproducibility and interpretability, hence scientific validity. In this paper, we suggest shifting the focus from the direct detection of socioeconomic status from raw images through image features, to the mapping of interpretable urban morphology of basic urban elements as an intermediate step, to which socioeconomic patterns can then be related. This shift is profound, in that, rather than abstract image features, it allows to capture the morphology of real urban objects, such as buildings and streets, and use this to then interpret other patterns, including socioeconomic ones. Because socioeconomic patterns are not derived from raw image data, the mapping of these patterns is less data demanding and more replicable. Specifically, we propose a 2-step approach: (1) extraction of fundamental urban elements from satellite imagery, and (2) derivation of meaningful urban morphological patterns from the extracted elements. We refer to this 2-step approach as “EO + Morphometrics”. Technically, EO consists of applying deep learning through a reengineered U-Net shaped convolutional neural network to publicly accessible Google Earth imagery for building extraction. Methods of urban morphometrics are then applied to these buildings to compute semantically explicit and interpretable metrics of urban form. Finally, clustering is applied to these metrics to obtain morphological patterns, or urban types. The “EO + Morphometrics” approach is applied to the city of Nairobi, Kenya, where 15 different urban types are identified. To test whether this outcome meaningfully describes current urbanization patterns, we verified whether selected types matched locally designated informal settlements. We observe that four urban types, characterized by compact and organic urban form, were recurrent in such settlements. The proposed "EO + Morphometrics" approach paves the way for the large-scale identification of interpretable urban form patterns and study of associated dynamics across any region in the world.

CORE (COnnecting REpositories)