Human-oriented Urban Design with Support of Multi-source Data and Deep Learning: A Case Study on Urban Greenway Planning of Suzhou Creek, Shanghai

As China’s urbanization drive enters the “second half”, increasing attention has been paid to the construction of high-quality living environment. Under this backdrop, urban greenways, an effective means to improve the spatial quality, have received high attention. However, the current greenway planning is mainly organized in a top-down macro perspective, which fails to effectively integrate human-based spatial and behavioral elements into the analysis framework. In this context, this study proposes a data-informed analytical approach combining classical urban design theories, multi-sourced urban data, and deep learning algorithm to compute the “where” and “how” questions of urban greenway planning, with the Suzhou Creek area of Shanghai as an example. According to the classical urban design theories, Cervero and Ewing’s 5Ds, i.e., density, diversity, design, dimensions of destination accessibility, and distance to transit, are selected as key factors. A series of new urban data, including street view images, points of interest (POI), location-based services (LBS) positioning data, structured web data and built environment data, are applied together with deep learning algorithms and geographical information system (GIS) tools to measure these key factors within the human-oriented resolution. Combining the constructability of existing streets, the suitability analysis of urban greenways is achieved via analytic hierarchy process and the street portraits are generated to assist detailed design decisions. This analytical approach is an attempt to add human-oriented concerns within a city-wide scale into the greenway planning. It also contributes to pushing the methodological boundaries of greenway planning by combining classical urban design thinking with new urban data and new techniques, which helps to assist human-oriented urban design practices.