A proposed framework for generating plausible cities

Journal Article (2026)
Author(s)

Oliver J. Post (TU Delft - Architecture and the Built Environment)

Hugo Ledoux (TU Delft - Architecture and the Built Environment)

Akshay Patil (TU Delft - Architecture and the Built Environment)

Research Group
Urban Data Science
DOI related publication
https://doi.org/10.5194/isprs-annals-XII-4-W1-2026-277-2026 Final published version
More Info
expand_more
Publication Year
2026
Language
English
Research Group
Urban Data Science
Journal title
ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Issue number
4/W1-2026
Volume number
12
Pages (from-to)
277-284
Event
21st International 3D GeoInfo Conference 2026 (2026-09-28 - 2026-09-29), Sofia, Bulgaria
Page Views
10
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Three-dimensional city models are widely used for simulation and analysis, yet their creation remains constrained by data availability, geometric errors, and the labour-intensive nature of manual modelling. Procedurally generated cities offer a practical alternative, but existing methods lack plausibility because they rely on generalised rule-based approaches rather than real urban data. In this paper, we propose the CityStack framework, a three-step pipeline for the analysis, encoding, and generation of plausible cities grounded in real-world morphological data. The framework decomposes urban form into hierarchical layers; in this paper, we focus on the minor road network and the building footprint layers. Using publicly available geospatial data from 43 cities worldwide, we introduce new morphological metrics for characterising street network patterns and apply a Gaussian Mixture Model to identify 13 global road typologies, alongside a supervised gradient boosting classifier that assigns one of 9 building footprint typologies. Urban form is encoded in a typology grid, a raster representation in which each 100 m × 100 m cell stores the dominant typology, serving as a compact, comparable intermediate format between analysis and generation. New city variants are synthesised by adapting typology grids via simulated annealing and generating roads and buildings from typology statistics. All components are implemented in an open-source command-line tool, making the full pipeline reproducible and accessible for a range of urban modelling applications.