Exploring the relationship between air temperature and urban morphology factors using machine learning under local climate zones

Abstract

Urban microclimate faces serious challenges due to increased urbanization and frequent heatwave events. Many studies focused on investigating the holistic quantitative relationships between urban morphology factors and heat island intensity at the city scale, but less effort has been devoted to exploring the relationships on a block scale. Additionally, there is a lack of fast prediction methods for urban microclimate for local climate zones (LCZ) planning and design. To address these challenges, this study proposes a Long Short-Term Memory Networks (LSTM) model to predict the effects of urban morphology factors on the air temperature under local climate zones. The effects of the spatial morphology features on the air temperature were characterized and quantified employing a post-interpretation method. The Pearl River New Town (PRNT), the downtown area of Guangzhou, China, was considered as the research area for the model implementation. The results showed that air temperature prediction accuracy is the best when using the historical three-time step data, with R2 of 0.975. LCZ A has the highest prediction accuracy, with an R2 of 0.990. LCZ 5 has the lowest accuracy, with an R2 of 0.881. Moreover, the effect of urban morphology factors on air temperature was found to be greater than the effect of land cover type. In this regard, the sky view factor (SVF) has the highest impact, followed by the aspect ratio (AR) and the pervious surface fraction (PSF). Nevertheless, the warming effect in built type was stronger than that in land cover. During the heatwave period, the maximum and minimum temperature changes were recorded in LCZ 4 and LCZ A, respectively, with values of 9.7 °C and 8.6 °C. It was shown that low-rise areas are more resilient than high-rise areas during heatwave periods. This is because low-rise areas generally exhibit a smaller increase in air temperature. These findings provide a better understanding of the relationship between urban microclimate and urban form, and a method of rapidly predicting the microclimate of a neighborhood block. It provides guidance and support, with great significance for climate-friendly urban planning.

Keywords

• Urban morphology
• Air temperature
• Urban microclimate
• Local climate zone
• Machine learning