文献类型：期刊文献

英文题名：Exploring the Non-Linear Impacts of Urban Features on Land Surface Temperature Using Explainable Artificial Intelligence

作者：Feng, Fei[1] Ren, Yaxue[2] Xu, Chengyang[1] Jia, Baoquan[3] Wu, Shengbiao[4] Lafortezza, Raffaele[1,2]

第一作者：Feng, Fei

机构：[1] Research Centre of Urban Forestry, Key Laboratory for Silviculture and Forest Ecosystem of State Forestry and Grassland Administration, Beijing Forestry University, Beijing, 100083, China; [2] Department of Soil, Plant and Food Sciences, University of Bari Aldo Moro, Via Amendola 165/A, Bari, 70126, Italy; [3] Research Institute of Forestry, Chinese Academy of Forestry, Beijing, 100091, China; [4] Future Urbanity & Sustainable Environment [FUSE] Lab, Division of Landscape Architecture, Department of Architecture, Faculty of Architecture, The University of Hong Kong, Hong Kong

年份：2023

外文期刊名：SSRN

收录：EI(收录号：20230372800)

语种：英文

外文关键词：Artificial intelligence - Forestry - Land surface temperature - Linear regression - Structure (composition) - Surface measurement - Surface properties - Urban growth - Vegetation

摘要：Extreme urban temperatures have emerged as crucial threats to urban ecosystems and sustainable urban development. Against this background, we developed a Random Forest (RF) model by means of eXplainable Artificial Intelligence (XAI) to examine the contributions of various impact features in regulating land surface temperature (LST) in the study area of Beijing, China. Multiple data sources were investigated, including LST, Normalized Difference Vegetation Index (NDVI), land cover, elevation, tree height, and building height. A grid of the study area (inner and outer cities), composed of 3416 boxes, 3x3 km, was used to extract mean values of impact features. RF outperformed the Multiple Linear Regression model (R2 of 0.89 vs 0.83) in predicting LST, demonstrating complex non-linear relationships between LST and impact features. By applying the eXplainable Artificial Intelligence method, our results suggest that the major impact features of LST in Beijing were elevation (44.19%), compactness of impervious surface (17.27%), NDVI (11.12%), proportion of impervious surface area (8.04%), and tree height (3.83%). The non-linear relationship between LST and impact features highlights the need for systematic planning of urban landscapes. This study provides state-of-the-art technology to gain novel insights into managing urban green spaces, and building development to mitigate hot urban environments. ? 2023, The Authors. All rights reserved.