Exploring city patterns globally: The intra-urban morphology through the scope of unsupervised learning

Abstract

Cities are complex systems with a unique composition of diverse elements and their relationships. Throughout history, humans form and shape cities by a range of functional, social, economical and political interactions. This diversity is reflected in the formation of individual built and non-built environments. However, similar elements and features form patterns that can be observed among multiple cities. City models try to understand the underlying processes that manifest into spatial patterns of urban form, but are often limited by a regional context and lack of comparable data. This master thesis aims to explore the urban morphology in a comparable framework on a global scale with the use of new consistent datasets such as the Local Climate Zones (LCZs) to describe the urban morphology of cities and the Morphological Urban Areas (MUAs) to delineate urban agglomerations. A search of urban morphological patterns is conducted without prior knowledge on subsets of 1523 cities. With state of the art methods of unsupervised learning 138 clusters of urban morphological patterns are found. The patterns show urban morphological configurations with similar statistical and spatial characteristics. A similarity metric is developed to compare cities based on the found patterns. Grouping similar cities leads to the formation of clusters which are partially congruent with geographic regions. The results of this work show that the formation of patterns with similar urban morphological configurations is linked to the geographic location. This master thesis is a first step towards a comprehensive knowledge on the formation of urban morphological configurations and contributes to a better understanding of cities.