Unsupervised clustering approach to residential typo-morphologies across multiple cities for urban heat vulnerability assessment

Journal Article (2026)
Author(s)

Maha Habib (TU Delft - Environmental Technology and Design)

Doruntina Zendeli (Politecnico di Milano)

Marjolein van Esch (TU Delft - Environmental Technology and Design)

Wim J. Timmermans (University of Twente)

Maarten van Ham (TU Delft - Urban Studies)

Research Group
Environmental Technology and Design
DOI related publication
https://doi.org/10.1016/j.scs.2025.107107
More Info
expand_more
Publication Year
2026
Language
English
Research Group
Environmental Technology and Design
Volume number
137
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

Residential environments are central to addressing urban heat stress for vulnerable populations and are prime target areas for implementing climate adaptation strategies. The reliance on urban heat island (UHI) intensity mapping alone has been argued to provide limited guidance for adaptation efforts, whereas linking heat patterns to the built environment characteristics through frameworks such as Local Climate Zones (LCZ) provides actionable insights for developing neighborhood cooling strategies. However, the widely used LCZ maps have a few limitations, such as misrepresenting variation within types because they cannot account for sub-classes beyond the standardized framework. This paper presents an unsupervised clustering approach to identify residential typo-morphologies across 99 Dutch cities, enhancing their relevance for urban heat vulnerability assessments. The analysis reveals that five morphological and canopy parameters (FSI, GSI, OSR, Havg, and FVC) selected from 17 parameters are sufficient to identify nine distinct residential typo-morphologies relatable to LCZs within 100 m × 100 m grid cells. The evaluations demonstrate that our approach detects underrepresented LCZ types and reveals new sub-classes absent from standard LCZ classifications. Key findings include detection of high-density areas (LCZ 42) reflecting recent urban densification with one of the highest UHImax next to LCZ 2 (4.2–4.9 K), and vegetation-differentiated variants within sparse and low-rise categories LCZ 9D and LCZ 6D, distinguished by distinctive UHImax (0.5–0.7 K) higher compared to their reference base types. Notably, tree coverage remains low across low-rise and compact typo-morphologies, revealing substantial opportunities for greening interventions. This data-driven refinement preserves LCZ's global comparability while considering local specificity, providing improved frameworks to inform targeted climate adaptation strategies in residential environments.