kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Analysing the effect of tree canopy and urban form on urban surface heat using machine learning
KTH, School of Architecture and the Built Environment (ABE), Sustainable development, Environmental science and Engineering, Sustainability Assessment and Management. (Environmental Management and Assessment)
KTH, School of Electrical Engineering and Computer Science (EECS), Computing and Learning Systems.ORCID iD: 0009-0000-5380-7949
Show others and affiliations
(English)Manuscript (preprint) (Other academic)
Abstract [en]

Urban heat island (UHI) effects are intensified by climate change and urbanisation, increasing risks to human health, energy consumption, and urban sustainability. Understanding how urban form and vegetation influence surface temperature at fine spatial scales is therefore essential for developing effective heat mitigation strategies. This study investigates the relationships among tree canopy, urban form, and surface heat patterns across central Stockholm using high-resolution opportunistic infrared (IR) measurements and machine learning techniques. Surface temperature data were collected during summer 2022 using mobile City Scanner sensors mounted on electric vehicles, providing detailed street-level spatial coverage. Surface heat was expressed as the difference between surface temperature and air temperature (ST–AT). Independent variables representing urban morphology, tree canopy, sun exposure, sky-view factor, and water bodies were analysed at multiple spatial scales. Machine learning models, including XGBoost and GPBoost, were applied to identify the most important predictors of surface heat patterns.

Results show that hyper-local variables, particularly sun exposure, building density, and tree canopy within approximately 15 metres, had the strongest explanatory power for ST–AT. Larger spatial scales did not substantially improve model performance. GPBoost demonstrated considerably better predictive accuracy than XGBoost, highlighting the importance of accounting for spatial relationships in modelling urban heat processes. Tree canopy exhibited a clear cooling effect at local scales, primarily through shading, while water bodies exerted limited influence due to their uneven spatial distribution. The study demonstrates the value of combining opportunistic sensing with spatial machine learning approaches to capture detailed urban heat patterns. The findings emphasise the importance of local-scale urban design strategies, particularly tree placement and urban morphology, in mitigating heat exposure within the urban canopy layer.

National Category
Environmental Sciences
Research subject
Land and Water Resources Engineering
Identifiers
URN: urn:nbn:se:kth:diva-382204OAI: oai:DiVA.org:kth-382204DiVA, id: diva2:2062140
Funder
StandUp
Note

QC 20260526

Available from: 2026-05-25 Created: 2026-05-25 Last updated: 2026-05-26Bibliographically approved
In thesis
1. Analysing the effect of tree canopy and urban form on urban surface heat using street IR imagery
Open this publication in new window or tab >>Analysing the effect of tree canopy and urban form on urban surface heat using street IR imagery
2026 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

Urban heat island (UHI) effects are intensifying due to climate change and urbanisation, posing increasing risks to human health, energy demand, and environmental sustainability. Understanding how urban morphology and vegetation influence thermal conditions at street level remains challenging due to the limited spatial resolution of traditional measurement approaches. This thesis investigates urban heat dynamics in Stockholm using opportunistic drive-by sensing (DS) combined with spatial machine learning methods to analyse hyperlocal relationships between urban form, tree canopy, and surface temperature within the urban canopy layer. High-resolution air and surface temperature data were collected during the summers of 2021 and 2022 using DS platforms mounted on electric vehicles, generating more than one million spatially distributed observations. These data were integrated with geospatial datasets describing urban morphology, greenery, and water bodies. Relationships between environmental variables and thermal patterns were analysed using statistical methods and machine learning models, including XGBoost and GPBoost, across multiple spatial scales. The results demonstrate that DS measurements capture substantial hyperlocal variability in surface temperature that is not fully captured by weather stations or satellite-derived land-surface temperatures. Surface temperatures were generally higher and more spatially heterogeneous than air temperature, reflecting strong dependence on immediate land cover and urban geometry. Tree canopy and reduced sun exposure were consistently associated with lower surface temperature differences, while higher building density increased heat accumulation. Machine learning models showed the highest explanatory power at a hyperlocal scale, with GPBoost outperforming XGBoost due to its ability to account for spatial dependencies. The findings highlight the importance of integrating fine-scale sensing with advanced spatial modelling to improve understanding of urban heat processes. The proposed methodological framework supports evidence-based urban planning strategies to enhance climate resilience through vegetation, shading, and urban design.

Abstract [sv]

Urbana värmeöeffekter (Urban Heat Island, UHI) förstärks i takt med klimatförändringar och urbanisering, vilket medför ökade risker för hälsa, energianvändning och urban hållbarhet. Trots omfattande forskning kvarstår betydande kunskapsluckor kring hur urban morfologi och vegetation påverkar temperaturförhållanden på detaljerad rumslig skala i det urbana landskapet. Denna licentiatavhandling syftar till att analysera urbana värmemönster i Stockholm med särskilt fokus på samspelet mellan bebyggelsestruktur, trädskikt och yttemperatur, genom användning av högupplöst mobil datainsamling och rumslig maskininlärning.

Studien baseras på mobil datainsamling (drive-by sensing), där temperaturdata samlats in med sensorer monterade på eldrivna fordon under somrarna 2021 och 2022. Datamaterialet omfattar över en miljon observationer av lufttemperatur och yttemperatur, vilka har integrerats med geodata om bebyggelse, vegetation och vatten. Samband mellan miljövariabler och temperaturmönster analyserades med statistiska metoder samt maskininlärningsmodeller, främst XGBoost och GPBoost, på flera rumsliga skalor.

Resultaten visar att DS-metoden möjliggör analys av betydande hyperlokal variation i yttemperatur, som inte fångas av traditionella väderstationer eller satellitbaserade observationer. Yttemperaturen uppvisar större variation och generellt högre värden än lufttemperaturen, vilket speglar stark påverkan från lokala markegenskaper och urban geometri. Trädskikt och minskad solexponering är konsekvent associerade med lägre temperaturdifferenser, medan hög bebyggelsetäthet bidrar till ökad värmeackumulering. Maskininlärningsmodellerna uppvisar högst förklaringsgrad på hyperlokal skala, där GPBoost presterar bättre än XGBoost genom att explicit beakta rumsliga beroenden.

Avhandlingen visar att kombinationen av högupplöst mobil miljömätning och avancerad rumslig modellering ger förbättrade möjligheter att förstå urbana värmeprocesser. Resultaten understryker vikten av finupplöst analys för att identifiera lokala värmemönster som är relevanta för mänsklig exponering. Den föreslagna metodansatsen bidrar till utvecklingen av evidensbaserade planeringsstrategier för klimatanpassning, där vegetation, skuggning och urban form utgör centrala komponenter för att minska värmebelastning i städer.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2026. p. 46
Series
TRITA-ABE-DLT ; 2618
Keywords
Urban heat island, surface temperature, drive-by sensing, machine learning, tree canopy, urban morphology, Urbana värmeöeffekter, yttemperatur, mobil datainsamling, maskininlärning, trädskikt, urban morfologi
National Category
Environmental Sciences
Research subject
Land and Water Resources Engineering
Identifiers
urn:nbn:se:kth:diva-382207 (URN)978-91-8106-639-5 (ISBN)
Presentation
2026-06-15, H1, Teknikringen 33, Floor 5, KTH Campus, public video conferenece link https://kth-se.zoom.us/j/63024958953, Stockholm, 13:00 (English)
Opponent
Supervisors
Funder
StandUp
Note

QC 20260525

Available from: 2026-05-25 Created: 2026-05-25 Last updated: 2026-06-01Bibliographically approved

Open Access in DiVA

No full text in DiVA

Authority records

Merdymshaeva, ElinaGhani, ShadaabLjungqvist, LukasHåkansson, AnneMörtberg, Ulla

Search in DiVA

By author/editor
Merdymshaeva, ElinaGhani, ShadaabLjungqvist, LukasHåkansson, AnneMörtberg, Ulla
By organisation
Sustainability Assessment and ManagementComputing and Learning SystemsArchitecture, Landscapes and Cities
Environmental Sciences

Search outside of DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric score

urn-nbn
Total: 29 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf