Existing studies have shown some relationships between the epidemic and the built environment during the COVID-19 pandemic. However, previous studies have paid more attention to the linear relationship but limited in exploring the non-linear relationship between them. This study first collected the confirmed case data of 431 communities in Wuhan urban center area, and constructed built environment variables based on a modified 5Ds theory with density, diversity, design, transit accessibility, and destination accessibility. The gradient boosting decision tree (GBDT) model is then applied to explore the nonlinear relationship and the threshold effect between the COVID-19 epidemic and built environment. The results show a strong nonlinear relationship between epidemic and built environment by comparing GBDT and OLS models. The density and design variables contribute most to the COVID-19 epidemic, especially population density, which presents the greatest relative importance (25.62%). Moreover, several the built environment variables have prominent threshold effects on the COVID-19 epidemic. Particularly, the effects of aging rate and shopping facilities on the COVID-19 epidemic do not show a consistent trend when these two variables exceed a certain threshold. These findings may provide several implications with practical value on how to effectively prevent and control the COVID-19 epidemic in high-density cities from the perspective of urban planning.

The COVID-19 pandemic; Build environment; Gradient boosting decision tree; Nonlinear relationship; Threshold effect