The Association of Urban Greenness and Walking Behavior: Using Google Street View and Deep Learning Techniques to Estimate Residents' Exposure to Urban Greenness.

Abstract

Many studies have established that urban greenness is associated with better health outcomes. Yet most studies assess urban greenness with overhead-view measures, such as park area or tree count, which often differs from the amount of greenness perceived by a person at eye-level on the ground. Furthermore, those studies are often criticized for the limitation of residential self-selection bias. In this study, urban greenness was extracted and assessed from profile view of streetscape images by Google Street View (GSV), in conjunction with deep learning techniques. We also explored a unique research opportunity arising in a citywide residential reallocation scheme of Hong Kong to reduce residential self-selection bias. Two multilevel regression analyses were conducted to examine the relationships between urban greenness and (1) the odds of walking for 24,773 public housing residents in Hong Kong, (2) total walking time of 1994 residents, while controlling for potential confounders. The results suggested that eye-level greenness was significantly related to higher odds of walking and longer walking time in both 400 m and 800 m buffers. Distance to the closest Mass Transit Rail (MTR) station was also associated with higher odds of walking. Number of shops was related to higher odds of walking in the 800 m buffer, but not in 400 m. Eye-level greenness, assessed by GSV images and deep learning techniques, can effectively estimate residents’ daily exposure to urban greenness, which is in turn associated with their walking behavior. Our findings apply to the entire public housing residents in Hong Kong, because of the large sample size.