Prediction of Urban Thermal Environment Based on Multi-Dimensional Nature and Urban Form Factors

Abstract

The urban thermal environment is affected by multiple urban form and natural environment factors; research on the accurate prediction of the urban thermal environment, considering the interaction among different urban environmental factors, is still lacking. The development of a machine learning model provides a good means of solving complex problems. This study aims to clarify the relationship between urban environmental variables and the urban thermal environment through high-precision machine learning models as well as provide scenarios of future urban thermal environment developments. We defined an urban thermal environment index (UTEI), considering twelve urban form and natural indicators sourced from the remote sensing data of 150 cities in the Jing-Jin-Ji region from 2000 to 2015. We achieved accurate predictions of UTEI through training a gradient-boosted regression trees model. By unpacking the model, we found that the contribution rate of elevation (ELEV) was the highest. Among all the urban form indicators, the elongation index (ELONG), urban population (POP), nighttime light intensity (NLI), urban area size (AREA), and urban shape index (SHAPE) also had high contributions. We set up five scenarios to simulate the possible impact of different urban form factors on the overall urban thermal environment quality in the region. Under extremely deteriorated patterns that do not control urban expansion and vegetation reduction, the average UTEI could be as high as 0.55–0.76 °C in summer and 0.24–0.29 °C in winter, yet in the extremely optimized situation, UTEI decreased by 0.69 °C in summer and 0.56 °C in winter. Results showed that better urban form improves the quality of urban environments and can provide important insights for urban planners to mitigate urban heat island problems.

Keywords

• urban thermal environment
• machine learning
• urban form
• feature importance
• scenario prediction