A machine learning framework for predicting potential urban cool island locations using Landsat 9 data: a case study of Oued Zenati, Algeria
Abstract
Urban heat stress is increasing in semi-arid Mediterranean towns, making the identification of Urban Cool Island (UCI) patterns essential for climate-resilient urban planning. This study developed an integrated remote sensing, GIS, and explainable machine-learning framework to map potential UCI locations in Oued Zenati, northeastern Algeria, using Landsat 9 imagery and geospatial environmental variables processed in Google Earth Engine. ArcGIS Desktop 10.8 was used for the final cartographic layout and map generation. Predictor variables included land surface temperature, albedo, elevation, slope, distance to roads, distance to rivers, and population density. Because field observations of UCIs were unavailable, proxy reference samples were generated from vegetation, moisture, and built-up indicators. Under the prevailing semi-arid conditions, NDWI values remained below − 0.06 across the study area, and all cool-zone reference samples were therefore derived from the vegetation criterion (NDVI ≥ 0.20). Four machine-learning algorithms (Random Forest, Gradient Tree Boosting, CART, and MaxEnt) were evaluated using a 70/30 training-testing split and independent validation. Gradient Tree Boosting achieved the highest discrimination ability (ROC-AUC = 0.916), whereas Random Forest provided the best threshold-dependent performance, with an accuracy of 0.900, Kappa of 0.800, Precision of 0.929, and F1-score of 0.897. Susceptibility maps showed that cooling potential is concentrated in vegetated and agricultural areas, while built-up and bare surfaces exhibit low cooling potential. SHAP analysis identified elevation as the most influential predictor, followed by slope, albedo, and land surface temperature, whereas distance to roads, distance to rivers, and population density exerted comparatively weaker effects. The proposed framework provides a robust, interpretable, and transferable approach for Urban Cool Island susceptibility mapping, supporting climate adaptation, green infrastructure planning, and sustainable urban development in semi-arid cities.
Related
- Integrated machine learning and AHP-based framework for spatial delineation of rainwater harvesting potential
- Predicting rate of penetration (ROP) with machine learning models: A physics-validated comparative case study of Gradient Boosting Regressor (GBR), Random Forest (RF), and K-Nearest Neighbours (KNN)
- Predicting Chemical Stressor Effects on Conjugative Plasmid Transfer Rates Using Machine Learning.
- Defining and predicting groundwater drought dynamics in a temperate maritime region using Standardised Groundwater Index (SGI) and ensemble machine learning
- Predicting freeway incident duration using machine learning
BibTeX
@article{naidja2026machine,
title = {{A machine learning framework for predicting potential urban cool island locations using Landsat 9 data: a case study of Oued Zenati, Algeria}},
author = {Amina Naidja and Dalal Farid and Djamal Bengusmia and Haythem Dinar and Nouh Rebouh and Imtiyaz Akbar Najar and Nadeem A. Khan},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-72875-3},
url = {https://www.nature.com/articles/s41598-026-72875-3},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A machine learning framework for predicting potential urban cool island locations using Landsat 9 data: a case study of Oued Zenati, Algeria
AU - Amina Naidja
AU - Dalal Farid
AU - Djamal Bengusmia
AU - Haythem Dinar
AU - Nouh Rebouh
AU - Imtiyaz Akbar Najar
AU - Nadeem A. Khan
PY - 2026
DA - 2026/10/03
JO - Scientific Reports
DO - 10.1038/s41598-026-72875-3
UR - https://www.nature.com/articles/s41598-026-72875-3
AB - Urban heat stress is increasing in semi-arid Mediterranean towns, making the identification of Urban Cool Island (UCI) patterns essential for climate-resilient urban planning. This study developed an integrated remote sensing, GIS, and explainable machine-learning framework to map potential UCI locations in Oued Zenati, northeastern Algeria, using Landsat 9 imagery and geospatial environmental variables processed in Google Earth Engine. ArcGIS Desktop 10.8 was used for the final cartographic layout and map generation. Predictor variables included land surface temperature, albedo, elevation, slope, distance to roads, distance to rivers, and population density. Because field observations of UCIs were unavailable, proxy reference samples were generated from vegetation, moisture, and built-up indicators. Under the prevailing semi-arid conditions, NDWI values remained below − 0.06 across the study area, and all cool-zone reference samples were therefore derived from the vegetation criterion (NDVI ≥ 0.20). Four machine-learning algorithms (Random Forest, Gradient Tree Boosting, CART, and MaxEnt) were evaluated using a 70/30 training-testing split and independent validation. Gradient Tree Boosting achieved the highest discrimination ability (ROC-AUC = 0.916), whereas Random Forest provided the best threshold-dependent performance, with an accuracy of 0.900, Kappa of 0.800, Precision of 0.929, and F1-score of 0.897. Susceptibility maps showed that cooling potential is concentrated in vegetated and agricultural areas, while built-up and bare surfaces exhibit low cooling potential. SHAP analysis identified elevation as the most influential predictor, followed by slope, albedo, and land surface temperature, whereas distance to roads, distance to rivers, and population density exerted comparatively weaker effects. The proposed framework provides a robust, interpretable, and transferable approach for Urban Cool Island susceptibility mapping, supporting climate adaptation, green infrastructure planning, and sustainable urban development in semi-arid cities.
SN - 2045-2322
ER -