Revealing the relationship between 2D/3D built environment and jobs-housing separation coupling nonlinearity and spatial nonstationarity

Transit-oriented development encourages metropolises to alleviate jobs-housing separation (JHS) by optimizing the built environment (BE). Researchers have found that BE exerts different effects on home- and work-oriented JHS, but their statistical models ignored the nonlinear and spatially nonstationary features of the relationship. In this study, we collected location-based service data to identify commuters and aggregated them to 188 metro station areas from a home or work orientation. We then defined a parameter β to measure JHS following the distance decay law, added three-dimensional (3D) indicators to the “Node-Place-Function” system to describe the BE, and applied the LightGBM-SHAP after multi-model comparison to learn and visualize their complex relationships. We identified three types of BE variables: 1) exhibiting important effects only on one orientation, 2) exhibiting similar effects on two orientations, and 3) exhibiting opposite effects on two orientations. Particularly, we designed a new visualization framework for SHAP that synergizes nonlinear dependency with the spatial distribution, which can provide threshold targets and spatial ranges for policy regulation simultaneously. For example, Hanyang should increase its shared bicycle density to 900 vehicles/km2 to meet the travel needs of residents, while Jianghan should provide affordable housing (<180,00 yuan/m2) to meet the housing needs of workers. Our findings aim to encourage planners to consider nonlinear determined thresholds in regulating actual spaces, which can directly support the refined urban management under limited resource condition.