Scholarlink

<p>Accurate estimation of Land Surface Temperature (LST) is essential for alleviating the negative effects of land use land cover (LULC) changes on the urban thermal environment. This study proposes an integration of deep learning and geospatial data analysis for predictive modeling of LST in&#xa0;the urban center of Quang Ngai, Vietnam – a region with very limited documents on its thermal behavior. We retrieve LST data from the thermal band (Band 10) of the Landsat-8 satellite in the dry season of 2022. Google Earth Engine is used as the platform for data retrieval. Topographical characteristics, LULC, urban morphological features, and proximity-based variables are employed as influencing factors. Deep Transformer Network (DTN), Deep Neural Network (DNN), and Convolutional Neural Network (CNN) are employed to explore the relationship between LST and its explanatory variables. Experimental results point out the suitability of DTN in geospatial modeling of LST in the study area with a coefficient of determination (R<sup>2</sup>) of 0.88, followed by DNN and CNN, which both achieve the R<sup>2</sup> of 0.86. In addition, feature analyses show that built-up density, distance to coastlines, bareland density, greenspace density, and distance to rivers are the most influential factors. To forecast the LST and the intensity of the urban heat island (UHI) effect in 2030, this study relies on an integration of the Cellular Automata (CA) and Artificial Neural Network (ANN) for LULC projection. The new framework is able to identify and predict the hot spots of intense heat stress in the study area. Hence, local authorities, urban planners, and architects can leverage the proposed approach in formulating effective strategies for UHI effect mitigation and sustainable urban development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning-Based Predictive Modeling of Urban Heat Stress Using Transformer Network, Deep Neural Network, and Convolutional Neural Network

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Nhat-Duc Hoang,
  • Van-Duc Tran

摘要

Accurate estimation of Land Surface Temperature (LST) is essential for alleviating the negative effects of land use land cover (LULC) changes on the urban thermal environment. This study proposes an integration of deep learning and geospatial data analysis for predictive modeling of LST in the urban center of Quang Ngai, Vietnam – a region with very limited documents on its thermal behavior. We retrieve LST data from the thermal band (Band 10) of the Landsat-8 satellite in the dry season of 2022. Google Earth Engine is used as the platform for data retrieval. Topographical characteristics, LULC, urban morphological features, and proximity-based variables are employed as influencing factors. Deep Transformer Network (DTN), Deep Neural Network (DNN), and Convolutional Neural Network (CNN) are employed to explore the relationship between LST and its explanatory variables. Experimental results point out the suitability of DTN in geospatial modeling of LST in the study area with a coefficient of determination (R2) of 0.88, followed by DNN and CNN, which both achieve the R2 of 0.86. In addition, feature analyses show that built-up density, distance to coastlines, bareland density, greenspace density, and distance to rivers are the most influential factors. To forecast the LST and the intensity of the urban heat island (UHI) effect in 2030, this study relies on an integration of the Cellular Automata (CA) and Artificial Neural Network (ANN) for LULC projection. The new framework is able to identify and predict the hot spots of intense heat stress in the study area. Hence, local authorities, urban planners, and architects can leverage the proposed approach in formulating effective strategies for UHI effect mitigation and sustainable urban development.