Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Exploring the spatio-temporal effects of urban thermal environment on the heat island using geographically interpretable machine learning

Hu, Junbiao LU (2025) In Master Thesis Series in Environmental Studies and Sustainability Science MESM02 20251
LUCSUS (Lund University Centre for Sustainability Studies)
Abstract
The Urban Heat Island (UHI) effect exacerbates environmental degradation, growing energy demand, and public health risks, especially in densely populated and vulnerable Mediterranean urban communities. This study investigates the spatial heterogeneity of the UHI effect in Barcelona from 2015 to 2023 by integrating multi-source geospatial data and interpretable machine learning techniques. Using Google Earth Engine (GEE) and XGBoost models trained on remote sensing data and sociodemographic indicators, the GeoSHAP method was applied to quantify the local importance of UHI drivers at a resolution of 500 m × 500 m. The results reveal strong spatial heterogeneity: bare soil exposure (ΔNDBAL) becomes the main stress in central and coastal... (More)
The Urban Heat Island (UHI) effect exacerbates environmental degradation, growing energy demand, and public health risks, especially in densely populated and vulnerable Mediterranean urban communities. This study investigates the spatial heterogeneity of the UHI effect in Barcelona from 2015 to 2023 by integrating multi-source geospatial data and interpretable machine learning techniques. Using Google Earth Engine (GEE) and XGBoost models trained on remote sensing data and sociodemographic indicators, the GeoSHAP method was applied to quantify the local importance of UHI drivers at a resolution of 500 m × 500 m. The results reveal strong spatial heterogeneity: bare soil exposure (ΔNDBAL) becomes the main stress in central and coastal areas, while the mitigating effects of vegetation (NDVI) and altitude decrease in compact urban cores. In contrast, in marginal areas such as Nobarris, increased ΔSHAP values ​​indicate latent heat risks, which are often overlooked in traditional thermal assessments. Rather than prescribing fixed interventions, the study provides spatially explicit insights to support equitable and resilient climate strategies. It demonstrates how explainable machine learning can inform more context-sensitive and justice-focused urban planning within a sustainability framework. (Less)
Please use this url to cite or link to this publication:
author
Hu, Junbiao LU
supervisor
organization
course
MESM02 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Urban heat island, Spatial heterogeneity, GeoSHAP, Interpretable Machine Learning, Environmental justince, Sustainability Science
publication/series
Master Thesis Series in Environmental Studies and Sustainability Science
report number
2025:047
language
English
id
9190326
date added to LUP
2025-10-15 11:57:59
date last changed
2025-10-15 11:57:59
@misc{9190326,
  abstract     = {{The Urban Heat Island (UHI) effect exacerbates environmental degradation, growing energy demand, and public health risks, especially in densely populated and vulnerable Mediterranean urban communities. This study investigates the spatial heterogeneity of the UHI effect in Barcelona from 2015 to 2023 by integrating multi-source geospatial data and interpretable machine learning techniques. Using Google Earth Engine (GEE) and XGBoost models trained on remote sensing data and sociodemographic indicators, the GeoSHAP method was applied to quantify the local importance of UHI drivers at a resolution of 500 m × 500 m. The results reveal strong spatial heterogeneity: bare soil exposure (ΔNDBAL) becomes the main stress in central and coastal areas, while the mitigating effects of vegetation (NDVI) and altitude decrease in compact urban cores. In contrast, in marginal areas such as Nobarris, increased ΔSHAP values ​​indicate latent heat risks, which are often overlooked in traditional thermal assessments. Rather than prescribing fixed interventions, the study provides spatially explicit insights to support equitable and resilient climate strategies. It demonstrates how explainable machine learning can inform more context-sensitive and justice-focused urban planning within a sustainability framework.}},
  author       = {{Hu, Junbiao}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master Thesis Series in Environmental Studies and Sustainability Science}},
  title        = {{Exploring the spatio-temporal effects of urban thermal environment on the heat island using geographically interpretable machine learning}},
  year         = {{2025}},
}