Framework of street grid-based urban heat vulnerability assessment: Integrating entropy weight method and BPNN model
Fei Guo,Ruwei Zheng,Jun Zhao,Hongchi Zhang,Jing Dong
DOI: https://doi.org/10.1016/j.uclim.2024.102067
IF: 6.663
2024-07-26
Urban Climate
Abstract:In the context of global warming, the increasing frequency of high temperature and heatwave events poses a significant threat to human health in cold cities, particularly those located in northern China. This study focuses on Tianjin, considering its unique geographical and climatic conditions. Utilizing the urban spatial unit delineated by the street network, known as the Street Grid (SG), we establish a comprehensive framework comprising 12 indicators to assess heat vulnerability, encompassing exposure, sensitivity and adaptability. The Heat Vulnerability Index (HVI) is synthesized using the Entropy Weight Method (EWM). Our findings reveal spatial disparities in vulnerability within Tianjin's central urban area. The Slight vulnerability level comprises the largest proportion, representing 35.91% of the area. Additionally, the proportions of vulnerability levels categorized as Low, Moderate, Strong, and Extreme are 20.45%, 27.78%, 9.97%, and 5.50%, respectively. Furthermore, we develop a HVI assessment model based on the Back-propagation Neural Network (BPNN), achieving high accuracy and reliability. We compared two different index integration strategies: additive aggregation and multiplicative aggregation. The final model's Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were 0.0320 and 0.0493, respectively, demonstrating high precision and reliability. The established framework offers novel insights and practical tools for rapid urban heat vulnerability assessment in cold regions, informing urban space management, policy formulation, disaster management, and sustainable development decision-making in Tianjin, thereby providing robust support for addressing the challenges posed by climate change.
environmental sciences,meteorology & atmospheric sciences