Authors: Haruto Nakamura, Priya Sharma, Liam O'Sullivan
Journal: Advances in Urban Informatics and Cyber-Physical Systems (AUICPS), ISSN 3087-4890
Citation: AUICPS 1(1), 2024-01-31.
Type: Original Research
Urban Heat Island (UHI) intensity poses significant challenges to urban sustainability and public health. This study develops a machine learning framework to predict surface UHI intensity using Landsat-derived land surface temperature (LST) and land use/land cover (LULC) data. We employed random forest (RF) and support vector regression (SVR) models trained on multi-temporal Landsat 8 imagery (2013–2023) for a rapidly urbanizing city. Results show that RF outperforms SVR with an R² of 0.89 and RMSE of 1.2°C. Key predictors include impervious surface fraction, vegetation index, and albedo. The model successfully captures spatial heterogeneity of UHI, with hotspots concentrated in industrial and high-density residential zones. Our findings demonstrate the efficacy of integrating remote sensing with machine learning for UHI prediction, offering a scalable tool for urban climate adaptation planning.
Urban Heat Island, machine learning, Landsat, land surface temperature, random forest, urban climate, remote sensing