Effect of urban morphology on air pollution distribution in high-density urban blocks based on mobile monitoring and machine learning

• Published in: Building and environment Vol. 219; p. 109173
• Main Authors: Huang, Chenyu, Hu, Tingting, Duan, Yusen, Li, Qingyu, Chen, Nan, Wang, Qi, Zhou, Mengge, Rao, Pinhua
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2022
• Subjects: High-density urban blocks; Machine learning; Mobile monitoring; Pollutant concentration; Semantic segmentation; Urban morphology; High-density urban blocks; Mobile monitoring; Semantic segmentation; Pollutant concentration; Urban morphology; Machine learning
• ISSN: 0360-1323; 1873-684X
• DOI: 10.1016/j.buildenv.2022.109173

It is essential to investigate the morphological factors that contribute to air pollution's spatial distribution using mobile monitoring data, and to regulate them at the urban planning level. However, mobile monitoring data are unstable and more difficult to model under real-world atmospheric circumstances. This work assesses the nonlinear relationship between spatial distribution of air pollutants and building morphological indicators in a high-density city based on mobile monitoring and machine learning. By conducting a vehicle-mounted mobile monitoring experiment, we establish spatial distribution data sets for PM2.5 and PM10 on three typical regions in Huangpu District, Shanghai. 9 indicators of urban morphology are derived, including green view index and sky view factor, using semantic segmentation and deep learning on street-view images. Correlation analysis demonstrates that the difficulty lies in implementing linear modeling methods. The performances of six machine learning algorithms for predicting the spatial variability of pollutants are compared. The result shows that neural networks have the highest performance for repidly predicting pollutant diffusion levels in conceptual designs. •Spatial distribution of pollutants is strongly related to urban morphology.•Long-term mobile monitoring of pollutants in high-density urban blocks has been conducted.•Semantic segmentation-based indicators such as green view index have been investigated.•Machine learning is better at identifying non-linear patterns in pollutants distribution.•An open neural network model has optimal performance and allows for ongoing improvement.