Deep Learning-Based analytic framework using comprehensive urbanization index for heat vulnerability assessment in urban areas

Abstract

The objective of this study was to propose an analytic framework for heat vulnerability based on the detailed spatial unit of a city. Data were analyzed for Daegu city, South Korea, during the period from June 1, 2018, to October 30, 2018. Four types of physical and environmental indicators were used: building density (BD), urbanization density (UD), green density (GD), and park density (PD). By synthesizing these indicators with the principal component analysis (PCA) technique, we created a comprehensive urbanization rate (CUR) index that contributes to the relative comparison of the degree of urbanization between different regions. To verify the generated index, the distribution of the CUR index was compared with heat, and a positive correlation was confirmed. Finally, a heat prediction model was proposed that included the CUR index as the input variable. To resolve the limitation of spatial autocorrelation in traditional linear regression, a deep learning algorithm was introduced. The final model had a root-mean-squared error (RMSE) value of 0.296. In the future, it is expected that the proposed CUR index and the developed heat prediction model can be utilized as an analytic framework for heat vulnerability within a city.