Although many studies investigate the association between land use and station ridership, few examine their nonlinear and moderating relationships. Using metro smartcard data in Shenzhen, we develop a gradient boosting decision trees model to estimate the relative importance of land use variables and their threshold and moderating effects on ridership. We found that station betweenness centrality has the largest predictive power, followed by employment density and commercial floor area ratio (FAR). Results suggest that employment density, commercial FAR, and aggregate residential density should be set at 40,000 jobs/km2, 2, and 77,000 persons/km2, respectively, for maximizing ridership. The moderating effects show that population densification is more effective at terminal stations, whereas the policies intensifying nonresidential use work better at middle stations. These findings help planners prioritize land use strategies, identify effective ranges of land use metrics, and propose land use guidelines adaptive to the network position of stations.
Bibliographical noteFunding Information:
This study was supported by the National Natural Science Foundation of China ( 41801158 ), the Shenzhen Municipal Basic Research Project (Free Exploration) ( JCYJ20180302153551891 ), and the Shenzhen Municipal Natural Science Foundation ( JCYJ20190808173611341 ).
© 2020 Elsevier Ltd
- Gradient boosting decision trees (GBDT)
- Interaction effects
- Land use
- Nonlinear effects
- Transit-oriented development (TOD)