Machine Learning in Urban Morphology and Urban Climate: Prospects and Applications

Quantitative descriptions of urban morphology enhance our understanding of urban systems' operation and evolution. In recent years, with the rapid development of the AI, the application of machine learning in urban research has become increasingly widespread. Current applications can be broadly categorized into two main types:

The first category utilizes machine learning to reveal nonlinear relationships between urban morphology and ecosystem services. For example, research examines how spatial morphological indicators of urban green spaces or blue-green infrastructure affect vegetation's cooling effect, carbon sequestration, flood mitigation and other ecosystem services (Sun et al., 2019; Wang et al., 2023). This type of research breaks through the limitations of traditional linear analysis and can capture complex urban environmental interactions.

The second category employs deep learning-based representation-learning methods (e.g., contrastive self-supervised encoders, graph auto-encoders, Vision Transformers) for urban morphology clustering (de-Miguel-Rodriguez et al., 2025; Dong et al., 2019; Kempinska & Murcio, 2019). Traditional methods of urban classification, based on morphological indicators, often suffer from information loss, spatial mismatches, and lack of robustness. Deep learning techniques for high-dimensional feature extraction and latent variable representation have been developed, improving the robustness of urban classification. These advanced methods significantly enhance the accuracy and reliability of urban classification.

In this presentation, I will share empirical research findings in both areas, including specific cases in which I have participated, and discuss future development directions and application potential of this field in urban climate research.

Reference:
de-Miguel-Rodriguez, J., Requena-Garcia-Cruz, M. V., Romero-Sánchez, E., & Morales-Esteban, A. (2025). Automated building typology clustering and identification using a variational autoencoder on digital land cadastres. Results in Engineering, 26, 105232. https://doi.org/10.1016/j.rineng.2025.105232
Dong, J., Li, L., & Han, D. (2019). New Quantitative Approach for the Morphological Similarity Analysis of Urban Fabrics Based on a Convolutional Autoencoder. IEEE Access, 7, 138162–138174. https://doi.org/10.1109/ACCESS.2019.2931958
Kempinska, K., & Murcio, R. (2019). Modelling urban networks using Variational Autoencoders. Applied Network Science, 4(1), 114. https://doi.org/10.1007/s41109-019-0234-0
Sun, Y., Gao, C., Li, J., Wang, R., & Liu, J. (2019). Quantifying the Effects of Urban Form on Land Surface Temperature in Subtropical High-Density Urban Areas Using Machine Learning. Remote Sensing, 11(8), Article 8. https://doi.org/10.3390/rs11080959
Wang, M., Li, Y., Yuan, H., Zhou, S., Wang, Y., Adnan Ikram, R. M., & Li, J. (2023). An XGBoost-SHAP approach to quantifying morphological impact on urban flooding susceptibility. Ecological Indicators, 156, 111137. https://doi.org/10.1016/j.ecolind.2023.111137